When the MOTU AVB line first came out in 2015 it was put through its paces by end-users in order to gather real-world spec's rather than to simply treat official documentation as gospel.
Turned out you could run signal through the 16A's ins and outs, for example, 16 or more times (I think; it could well have been more) in-a-row before even the slightest degradation could be sensed. We're talking minuscule differences, even after that unrealistic loop-back.
Well-written plugins employ oversampling in order to retain signal integrity and it works.
IMHO the greatest destroyer of signal integrity is the end-user of the gear. DAW mixing with signals too-hot, employing the old-school M.O. of maximising signal strength in order to maximise the ratio to the noise floor, is IMHO the #1 mistake peeps make.
It's a different paradigm now. 24-bit means a ridiculous dynamic range so there's no need to max out your individual tracks, either during recording or mixing. Doing so in fact makes the result (mix) sound small, lifeless and choked in-comparison.
You mentioned compression a couple o' times. I'm guessing you meant data compression whereas the members here assumed you meant amplitude compression. If you were talking about the data version, I reckon you're correct - only FLAC delivers back what you started with. 320k MP3, if encoded by something like L.A.M.E., is plenty-fine for most peeps, including myself. iTunes encoding however produces a somewhat-smeared, lower-fidelity result at 320k due to Apple's decision to appease the masses by shortening the encode time; it's many times quicker than, say, L.A.M.E. and it's super-obvious that it's inferior 'cause the lack of HF content is immediately-obvious.
If you were talking about amplitude compression it's irrelevant 'cause bad compression usage and the rules around good practice apply equally to the analogue and digital domains.