Pafi: Yes i know the old usual way of solving the THD problem: include everything you want to hide in the feedback loop, and the output signal will look nice on paper.
This is not only found in UcD, but in many other platforms as well.
Let's call this: the easy solution...
Anyway i have found even if you use this method, it can only help you so much.
Let me make an analogy, which i think many have thought of or tried at some point in their carreer as Audio DIY'er: You want to build a class B amplifier without BIAS, and to make good THD figures you use a lot of feedback to get rid of the cross over distortion. It works in theory, but will help you very little in the real world. It will still sound terrible, no matter how much feedback you use, right?
In this analogy the cross over distortion is the hysteresis distortion in the ferrite core. You can make it look nice on paper, by using feedback, but you can never make it to sound good, if there is a lot of unlinear distortion in the coil. That's why you have to use a low distortion coil
even if you are using post filter NFB!
If you have an unlinear element in the feedback loop you can still get nice THD data using feedback, but the circuit will also act as a mixer, (like found in a radio input stage). A mixer blends tones together, and makes harmonics. So you can never get very good IMD, and the sound will never be any better, than without the feedback loop alltogether. Now i will say that UcD in particular has a quite low level of IMD, but that's not because of the feedback, it's more owing to the use of a good ferrite material in their output coil. Others are not taking so much consideration, and use sometimes very poor coils. As long as the feedback clears the THD, everything is good right ?
IMO True high performance demands a whole different approach.