The assumption that the layers are all cylindrical is a good first approximation.
The assumption that the layers form a logarithmic
spiral is not a good assumption at all, because it supposes that the
thickness of the paper at any point is proportional to its distance
from the center. This seems to me to be quite absurd.
An alternative assumption is that the layers form an Archimedean spiral.
This is slightly more realistic, since it says the paper has a uniform
thickness from beginning to end. But this assumption is not a much more
realistic than the assumption that all layers are cylindrical;
in fact, in some ways it is less realistic.
Here's how a sheet of thickness $h$ actually wraps around a cylinder.
First, we glue one side of the sheet (near the end of the sheet)
to the surface of the cylinder. Then we start rotating the cylinder.
As the cylinder rotates, it pulls the outstretched sheet around itself.
Near the end of the first full rotation of the cylinder, the
wrapping looks like this:
Notice that the sheet lies directly on the surface of the cylinder,
that is, this part of the wrapped sheet is cylindrical.
At some angle of rotation, the glued end of the sheet hits the part of
the sheet that is being wrapped. The point where the sheet is tangent to
the cylinder at that time is the last point of contact with the cylinder;
the sheet goes straight from that point to the point of contact with
the glued end, and then proceeds to wrap in a cylindrical shape around
the first layer of the wrapped sheet, like this:
As we continue rotating the cylinder, it takes up more and more layers
of the sheet, each layer consisting of a cylindrical section going
most of the way around the roll, followed by a flat section that joins
this layer to the next layer. We end up with something like this:
Notice that I cut the sheet just at the point where it was about to
enter another straight section. I claim (without proof) that this
produces a local maximum in the ratio of the length of the wrapped sheet
of paper to the greatest thickness of paper around the inner cylinder.
The next local maximum (I claim) will occur at the corresponding
point of the next wrap of the sheet.
The question now is what the thickness of each layer is.
The inner surface of the cylindrical portion of each layer of the
wrapped sheet has less area than the outer surface, but the portion of
the original (unwrapped) sheet that was wound onto the roll to make this layer had equal area on both sides. So either the inner surface was
somehow compressed, or the outer surface was stretched, or both.
I think the most realistic assumption is that both compression and stretching
occurred. In reality, I would guess that the inner surface is compressed more than the outer surface is stretched, but I do not know what the
most likely ratio of compression to stretching would be.
It is simpler to assume that the two effects are equal.
The length of the sheet used to make any part of one layer of the roll
is therefore equal to the length of the surface midway between the
inner and outer surfaces of that layer.
For example, to wrap the first layer halfway around the central cylinder
of radius $r$, we use a length $\pi\left(r + \frac h2\right)$
of the sheet of paper.
The reason this particularly simplifies our calculations is that the
length of paper used in any part of the roll is simply the area of the
cross-section of that part of the roll divided by the thickness of the paper.
The entire roll has inner radius $r$ and outer radius $R = r + nh$,
where $n$ is the maximum number of layers at any point
around the central cylinder. (In the figure, $n = 5$.)
The blue lines are sides of a right triangle whose vertices are
the center of the inner cylinder and the points where the first layer last touches the inner cylinder and first touches its own end.
This triangle has hypotenuse $r + h$ and one leg is $r$, so the other
leg (which is the length of the straight portion of the sheet)
is $$ \sqrt{(r + h)^2 - r^2} = \sqrt{(2r + h)h}.$$
Each straight portion of each layer is connected to the next layer
of paper by wrapping around either the point of contact with the glued
end of the sheet (the first time) or around the shape made by
wrapping the previous layer around this part of the layer below;
this forms a segment of a cylinder between the red lines with center at
the point of contact with the glued end.
The angle between the red lines is the same as the angle of the blue
triangle at the center of the cylinder, namely
$$ \alpha = \arccos \frac{r}{r+h}.$$
Now let's add up all parts of the roll. We have an almost-complete
hollow cylinder with inner radius $r$ and outer radius $R$,
missing only a segment of angle $\alpha$. The cross-sectional area of this is
$$ A_1 = \left(\pi - \frac{\alpha}{2} \right) (R^2 - r^2).$$
We have a rectangular prism whose cross-sectional area is the product
of two of its sides,
$$ A_2 = (R - r - h) \sqrt{(2r + h)h}.$$
Finally, we have a segment of a cylinder of radius $R - r - h$
(between the red lines) whose cross-sectional area is
$$ A_3 = \frac{\alpha}{2} (R - r - h)^2.$$
Adding this up and dividing by $h$, the total length of the sheet
comes to
\begin{align}
L &= \frac1h (A_1+A_2+A_3)\\
&= \frac1h \left(\pi - \frac{\alpha}{2} \right) (R^2 - r^2)
+ \frac1h (R - r - h) \sqrt{(2r + h)h}
+ \frac{\alpha}{2h} (R - r - h)^2.
\end{align}
For $n$ layers on a roll, using the formula $R = r + nh$,
we have $R - r = nh$, $R + r = 2r + nh$,
$R^2 - r^2 = (R+r)(R-r) = (2r + nh)nh$,
and $R - r - h = (n - 1)h$.
The length then is
\begin{align}
L &= \left(\pi - \frac{\alpha}{2} \right) (2r + nh)n
+ (n - 1) \sqrt{(2r + h)h}
+ \frac{\alpha h}{2} (n - 1)^2\\
&= 2n\pi r + n^2\pi h
+ (n-1) \sqrt{(2r + h)h}
- \left( n(r + h) - \frac h2 \right) \arccos \frac{r}{r+h}\\
&= n (R + r) \pi
+ (n-1) \sqrt{(2r + h)h}
- \left( n(r + h) - \frac h2 \right) \arccos \frac{r}{r+h}.
\end{align}
One notable difference between this estimate and some others
(including the original) is that I assume there can be at most
$(R-r)/h$ layers of paper over any part of the central cylinder,
not $1 + (R-r)/h$ layers.
The total length is the number of layers times $2\pi$ times the
average radius, $(R + r)/2$, adjusted by the amount that is missing in the
section of the roll that is only $n - 1$ sheets thick.
Things are not too much worse if we assume a different but uniform ratio
of inner-compression to outer-stretching, provided that we keep the
same paper thickness regardless of curvature; we just have to make an
adjustment to the inner and outer radii of any cylindrical segment of the roll, which I think I'll leave as "an exercise for the reader."
But this involves a change in volume of the sheet of paper.
If we also keep the volume constant, we find that the sheet gets thicker
or thinner depending on the ratio of stretch to compression and
the curvature of the sheet.
With constant volume, the length of paper in the main part of the
roll (everywhere where we get the the full number of layers) is the
same as in the estimate above, but the total length of the parts of the
sheet that connect one layer to the next might change slightly.
Update: Per request, here are the results of applying the formula
above to the input values given as an example in the question:
$h=0.1$, $R=75$, and $r=25$ (inferred from $R-r=b=50$), all measured
in millimeters.
Since $n = (R-r)/h$, we have $n = 500$.
For a first approximation of the total length of paper,
let's consider just the first term of the formula. This gives us
$$
L_1 = n (R + r) \pi = 500 \cdot 100 \pi \approx 157079.63267949,
$$
or about $157$ meters, the same as in the example in the question.
The remaining two terms yield
\begin{align}
L - L_1
&= (n-1)\sqrt{(2r + h)h}
- \left( n(r + h) - \frac h2 \right) \arccos\frac{r}{r+h} \\
&= 499\sqrt{50.1 \cdot 0.1} - (500(25.1) - 0.05)\arccos\frac{25}{25.1} \\
&\approx -3.72246774.
\end{align}
This is a very small correction, less than $2.4\times 10^{-5} L_1$.
In reality (as opposed to my idealized model
of constant-thickness constant-volume toilet paper), this
"correction" is surely insignificant compared to the uncertainties of
estimating the average thickness of the paper in each layer of a roll
(not to mention any non-uniformity
in how it is rolled by the manufacturing machinery).
We can also compare $\lvert L - L_1 \rvert$ to the amount of paper that
would be missing if the paper in the "flat" segment of the roll were
instead $n - 1$ layers following the curve of the rest of the paper.
The angle $\alpha$ is about $0.089294$ radians (about $5.1162$ degrees),
so if the missing layer were the innermost layer, its length would be
$25.05 \alpha \approx 2.24$, and if it were the outermost layer
it would be $74.95 \alpha \approx 6.69$ (in millimeters).
Just for amusement, I also tried expanding $L - L_1$ as a power
series around $h = 0$ (with a little help from Wolfram Alpha).
(To make $L - L_1$ a function of one variable $h$ with constants $R$ and $r$,
make the substitution $n = (R - r)/h$.)
This turns out to be a series of powers of $\sqrt h$ whose leading term is
$$
-\frac{(R + 2r)\sqrt2}{3\sqrt r} \sqrt h.
$$
Plugging in the values from the example, this evaluates to
approximately $-3.7267799625$.
If you really wanted the length of the idealized toilet roll to the
nearest millimeter, but could tolerate an error of a few $\mu\mathrm m$
(for typical dimensions of a toilet roll),
a suitable approximation would be
$$
L \approx \frac{\pi (R^2 - r^2)}{h} - \frac{(R + 2r)\sqrt2}{3\sqrt r} \sqrt h.
$$
