Al poly ESR should be at least ballpark right. I'm not aware that it changes much (unlike electrolytics which vary with temperature).
Although the aging mechanism is, I think, moisture seeping into the polymer, and then what, decomposing it, I don't know; does it lower conductivity so ESR rises, or does it act to disconnect areas so ESR rises while C falls, roughly proportionally?
Now; you've plotted the transfer function v2/v1, but that's not necessarily what we want. (But, again, it's not clear what's needed, lacking a model of the power converter.) Traditionally we measure the response of a filter as the insertion loss. For that, we use V1 = 2, Rs = 50, and first test with an RL = 50. Which is of course a simple 50% resistor divider, so we measure 1V at the load. Then we put the filter between Rs and RL, and measure the output voltage again (not the gain; or, half the gain to Vs(O/C), if you like).
The impedance of, like, the last bit, 2.2uH to 44uF, is 0.22 ohms, but ESR is 0.015 ohms, so the Q there could be quite high. ESR closer to 0.2 would be better.
Polymers are available in a range of ESRs; generally lower ESRs than tantalums (the staple of "bulk cap with known ESR"), but they have a lot of overlap; this should be doable.
The 1uH + 990uF is very low indeed, of course, but 3m ohm ESR is quite low, too, in fact Q = 1 falls at merely 10nH. So 1uH is a lot higher, though it's only noticeable against very low source impedances.
The fact that all these values are fairly mismatched, makes it inconvenient to try and force it into a prototypical filter; really I should be looking for the most important loops. The source end should be inconsequential -- it rolls off with L1 (sans the 47R/12R pad) at 8MHz. So we can ignore L1 at lower frequencies. (A good hint that you may want another cap in front of it -- one of those 330uF's perhaps?)
The next loop is the 330's to L2 to the 22's. This has a total equivalent of 44uF + 990uF = 42uF [1], so resonates with Zo = 0.23 ohm at Fo = 16.5kHz. Loop ESR is 38m ohm, so Q ~ 6.
[1] Capacitors in series ("+") use the parallel formula... I have no notation for "series" other than the plus sign.

When you have a resonant loop, any tap along it acts in a similar way; in effect we have a cap divider, which in turn acts as an RF impedance transformer, and we have a super low impedance on the left, and a modest impedance on the right. It acts parallel resonant, so we expect an impedance peak at V2 of around 0.23 * 6 = 1.4 ohms (at Fo).
You can measure port impedance by setting the other sources to zero, and placing a current source from GND to V2, of magnitude 1. V(V2) = (1A) * Z(V2) = Z(V2), so you read off ohms directly.

Since there's an impedance peak there, we might seek to shunt it; but we know better than to simply add more capacitance, because we'll just keep doing the same thing over again. If we use an R+C, with R ~ 0.2 ohm, and C >> 44uF (typically >= 2.5 times more), we'll end up with a deliciously flat impedance over the transition band.

You might also consider moving over C1 like I hinted earlier, then changing C3 to a damper (more ESR and C).
Back to the load again -- do you have any insight into it, at this time? Is it going to have a bunch of input capacitance like most converters? If so, we can model that; we could even model the converter's input resistance, which might range from positive to negative, depending.
If it's cap-input, consider putting a choke towards it. Which basically swaps around the filter, which solves the not-terribly-useful-L1 situation, and also gives you something to reason about, if you choose to shorten the filter instead (honestly, I'd be surprised if more than a CLC is needed, unless this is super sensitive, MIL or something).
Don't forget that common mode is often a worse offender than differential. You've got a damn fine filter here for differential, but it's all for naught if there's a 100MHz spike blasting straight over that common ground wire!
Tim