Interesting question. I try my 2 cents on the matter, it's just a guess, don't take me seriously.
In a BJT, saturation voltage \$V_{CE}\$ depends on many factors, the two most important being the voltage difference between the two forward biased junctions (\$V_{BE}-V_{BC}\$) and the voltage drop across the parasitic resistances, specially the collector one.
In your case collector current is minuscule, so probably the collector parasitic resistance is not important.
Voltage difference between BE and BC can be derived from Ebers-Moll or transport equations and the result is (Millman Grabel - Microelectronics)
$$ V_{CE_\text{sat}}=V_T\ln \left(\frac{\frac{1}{\alpha_R}+\frac{\beta_\text{for}}{\beta_R}}{1-\frac{\beta_\text{for}}{\beta_F}}\right ) $$
where \$\beta_\text{for}\$ is the forced beta, that is the ratio between collector current determined by the load and base current injected into the base (in your data sheet it's 10).
\$\beta_F\$ and \$\beta_R\$ are the forward and reverse current gains. \$\alpha_R=\frac{\beta_R}{\beta_R+1}\$
is the reverse base transport factor.
Betas aren't fixed numbers for a transistor, carved in stone, they vary with current, temperature, collector voltage, moon phase... (see for an in dept analysis of beta changes: Gray Meyer - Analysis and Design of Analog Integrated Circuits)
I suspect that in this case when the tiny \$I_C\$ increases, \$\beta_F\$ increases as well (in theory it should be proportional to \$\sqrt{I_C}\$), the logarithm argument decreases and \$ V_{CE_\text{sat}}\$ decreases. Unfortunately I have no news about the behavior of \$\beta_R\$ :(
A final general warning about data sheets, not really applying to this case. "Static parameters", like saturation voltage, beta..., are measured with a pulsed technique to avoid heating the silicon. Unfortunately at very low current there is no time to charge or discharge the parasitic capacitances and in some cases this effect is even visible on data sheets.