It's a peculiar thought, to consider a power supply as a two-port:
simulate this circuit – Schematic created using CircuitLab
A power supply is generally a very nonlinear system, particularly since we're talking switching power supplies. So it's not a very useful or meaningful analysis, right off the bat.
It's not a concept without merit, though. We can intentionally design an SMPS stage to act as a DC transformer (a somewhat esoteric, but illustrative example: Dave Berning's "ZOTL" tube amplifiers), or a mixer (class D amplifiers -- not too different from audio examples -- can be used to transmit AM radio directly, with a suitable output matching/filtering network), and though most (controlled or regulated) power supplies will have a complicated (nonlinear, history-dependent) response, we can still perform linearization around a point and come up with some input-output transfer function, relating input and output voltage and current. That is: a two-port.
For a typical case like a DC-DC buck converter with output voltage regulation, we'll still have the case that, for a step change in load current, say, the input current changes proportionally, albeit delayed by the input and output filters, and the controller's transfer function.
So, we can generalize such converters just a little bit, and consider them as pass-thru devices, for frequencies below their bandwidth; not exactly straight-line pass-thru I mean, but as a transformer, because energy must be conserved (minus losses). As for the apparent lowpass response -- it's clear that such a converter is also an energy-storage element. Which is pretty apparent from its design (an energy-storage element [inductor] is the active ingredient!), but the exact way that element manifests depends on the topology and current operating state (i.e. it's very different for buck/boost, DCM, CCM, resonant, take your pick; the input-output transfer function need not be in any obvious proportion to any particular component values in the circuit, but will be a function of many things all at once).
Usually, the transfer function is dominated by the input and output filters, and the control loop bandwidth is somewhere above the filter cutoff but below the switching frequency.
Can we generalize further to AC converters?
Well, sort of.
Since the input is rectified*, obviously we're not going to be delivering power to the AC line. At least, not more than a modest change or difference on top of baseline consumption. And, if it doesn't have PFC (the unit pictured does not), what gets to the line will be further gated by conduction of the rectifier itself (which is intermittent, hence the low power factor of this configuration).
*The exception being grid-tied inverters, when made to alternately sink or source power from the mains. That is, synchronous converters will be used throughout: either a direct mains chopper for PFC+inverter, or a PFC (say, normal boost style or whatever) in parallel with an inverter, and some means to bypass the input rectifier. Whatever it is, it's just a more messy realization of the simplest (four-quadrant inverter) case; the same thing "with extra steps".
But most of all, the power supply needs hold-up time between successive mains peaks. This is what the big electrolytics provide, and they represent (again, in the hand-waving linearized sense I'm talking within here) a pole to the transfer function, a quite low one of some 100s or 10s of Hz, and so we don't expect to pass signals much above this point.
The control loop anyway may have quite a low cutoff; the pictured supply appears to be a voltage-mode half-bridge forward converter, which for a number of reasons, usually ends up with a pretty low cutoff frequency (low ~kHz).
Why would we care?
Good question. Is the concern about mains emissions? We can apply the framework developed (hinted at?) above to answer it.
Given that an SMPS must already have adequate (input and output) filtering to deal with its own switching noise (of amplitude comparable to nominal output), and that that switching noise is typically in the 100kHz range (and harmonics), and given that most regulations limit the emissions above 150kHz (and either don't restrict emissions at all below that, or give a much more generous margin on it), we can simply conclude it's extremely likely, or outright impossible, to abuse the output of a power supply in such a way that output switching noise is effectively conducted (in the above sense, viewing the supply as a transfer function), to the mains, differentially.
The above considers emissions only, of course (or, transmission in general, as of the two-port).
There may be practical, or indeed functional, reasons to limit the emissions into a power supply itself, regardless whether it will transmit those emissions elsewhere (say to a regulated mains port).
As a rule, I would suggest avoiding RMS ripple currents more than, say, 1/10th the DC rating of the supply. Most supplies will use electrolytic capacitors at the output, rated to handle their own switching ripple, and not much more than that. Some supplies may use very little ripple (and rating) indeed -- voltage-mode converters (like this one possibly is) can be one, and resonant/LLC converters can be another. The practical/functional limitation here, then, would be not overheating or exploding those capacitors.
At lower frequencies (below the supply's control loop cutoff), capacitor ripple will be traded off to the primary side (thus, the load will be borne by the input caps), or indeed mains itself, but 5kHz (plus harmonics) you'll probably want to handle the lion's share of it locally. This also reduces RMS current, and AC voltage drop, on the wiring between supply and your circuit.
As for your filter network, I wouldn't bother with explicit ESR; merely choose electrolytics of large enough value, and ESR, suitable to dampen the CLC network formed around the inductor.
You might keep the cutoff of that CLC at a middle frequency, so that the pulsed load is borne by all the electrolytics together (they act in parallel), while higher frequencies are confined locally. That is, say Fsw ~ 330kHz and Fpulse ~ 5kHz, putting Fc ~ 40kHz gives on the order of 50dB attenuation at 330kHz+, while largely sharing current at lower frequencies.
I'd have to run some numbers (and I'd suggest you do so yourself, just as well -- indeed, you have the power supply handy, so you can look up what capacitors it actually uses at the output, and model them, plus wiring inductance between supply and your circuit, and your filter), but I suspect your capacitors are much too small. The supply itself probably uses high 100s to low 1000s uF on its output, and you'll likely need to use about as much to keep RMS currents within ratings.
Buck Converter Schematic:
Output Waveforms:
