Add to the above that as you add more NFB to a circuit, the ratio of higher order to lower order harmonics increases. So at 40 db NFB, you might get more 5th and 7th harmonic distortion relative to 2nd harmonic distortion, though total THD will decrease. Higher harmonics are generally considered less desirable because they have a less musical relationship to the input signal.
This article by Nelson Pass may be of interest:
https://www.passdiy.com/pdf/distortion_feedback.pdf
If I'm reading that paper right, Figure 10 seems to say that going from 0 to 15 dB feedback to gives you a higher proportion of higher harmonics, but going from there to 40 db makes all of them VASTLY lower and presumably entirely masked by the overwhelmingly clean signal (except possibly for the 2nd harmonic, which is comparatively benign).
Most of the curves are nonmonotonic, all but the 2nd harmonic rising between 0 dB feedback and 7 to 15 dB feedback, depending on the harmonic number, then all decline monotonically.
Figure 11 emphasizes that weirdness before the monotonic decline, giving distortion curves for 3 to 15 dB feedback, i.e., corresponding to the problematic region of Figure 10, where harmonics are rising because you're just not using enough feedback.
As he says in the text Figure 11 "clearly shows the increase in higher order harmonics with the application of negative feedback," but he doesn't address what he just said with regard to the nonmonotonicity in Figure 10, that "Negative loop feedback creates higher order distortion harmonics, and there seems to be an implication that you might want to use lots of feedback if you plan on using any at all."
Well, YEAH. It sure does seem from Figure 10 that if you're going to use feedback, you want to do it right and use enough feedback to get distortion really low. After noting the obvious implication that you should use enough feedback to do the job, he says "Some designers look at it this way, others use feedback sparingly, and some refuse to use it at all."
Whut? He doesn't explain why the designers who "look at it this way" aren't simply right, and the designers who use feedback more sparingly aren't simply missing the boat.
It seems to me that the main lesson of that paper should be that you shouldn't cascade multiple stages that generate significant distortion, because that increases the ugly higher-order harmonics and IM distortion, so you should NOT use a tube mic AND a tube preamp AND a tube compressor, and with transformers all over the place. You should generate the distortion you want in ONE stage (if desired), probably applied to individual instruments rather than more complex signals, and everything past that should be clean.
