Some of these graphs of THD+noise vs power point to why I don't like that spec. I prefer to look at noise and look at THD. The typical shape we see above usually indicates THD is staying rather low, and is being swamped by noise as the signal level goes down. I'd prefer a 3D plot showing THD, noise and power on 3 axis.
If you look at the same graph for the Denon it is better than any of these other graphs posted as it stays almost flat into lower powers, and eventually doesn't rise as much. That is because the Denon is relatively low in distortion, but much lower in noise than the others pieces of gear. We can hear into noise some 10-20 db. So the spectrum of the noise vs the signal will mean two similar plots at low signals may sound a bit different.
Optimum of course is pairing low thd with low noise.
Here as a point of reference is a Bel Canto Black as measured by Stereophile showing that switching amps can do well.
The Bel Canto Ref 600 mono which stays below .01% until past 200 wpc. It has a bit different shape on the THD+N curve than usual.
Now these are not inexpensive nor even necessarily the best that can be had. Just thought it worth showing what is possible.
Here is an earlier IcePower based Bel Canto. It stays below .1 % from lowest powers to highest powers.
Here is a Mark Levinson 536 which is a class AB amp which stays below .02% at 400 wpc and is .004% at 1 watt. Because it low noise and low distortion.
And even better at much less cost (though less power) is the Benchmark Amp. Never worse than .0005% and mostly around .0001%.
All these measurements from Stereophile testing.
So I'd like to see an amp have such a curve that is .01% or less at all levels up to rated power. Plenty have low enough THD, but many affordable amps don't have low enough noise. Such a curve at .1% and less might be inaudible, but having the extra 20db room for certainty is very possible with modern amps.