OK, well that was a bit of a waste of time. I picked a random amplifier here, a nice Rotel RA-970BX. Rated 80wpc@8R THD<0.03% at full power.
View attachment 181210
Its THD was less than 0.002% at 80wpc and it will put out 100wpc. (200kHz bandwidth) The signal source has an inherent THD of ~0.0015% itself.
To match the Benchmark 'test' I tried a 282mV (0.01W) output into 8R, but the noise with respect to that output is too great and the analyzer residual shows nothing useful, but certainly no X/over distortion.
At 1V@8R output, the distortion is completely buried in the noise on the output of the THD analyzer. There are no crossover spikes and even amplifying the residual and running a 1024 averaging on the scope shows nothing of note- no spikes. I backed the bias right down, but the range is such that there is still enough bias (and NFB) to essentially leave everything still buried in the noise floor.
This is the distortion residual amplified by 1000 times (60dB) with a 1024 averages to kill the noise.
View attachment 181209
View attachment 181206
I could say the THD+N at 2.83V@8R output is (left) 0.047% and (right) 0.044%, but it is all N. I cannot see on either scope any hamonics at all and certainly no crossover distortion. I have no reason to doubt Rotel's spec as being 250mW to rated output.
So I need an amplifier with an extremely low residual noise floor, then kill the bias altogether, view the spikes and send the analyzer output after the notch to the FFT.