In short, the imd hump doesn't really only affect low volume. It actually only indicates there is issues about modulation distortion.
More verbose: The reason why (I have returned the unit) it has a mid volume hump is probably the weird modulation shifts, and it shifts inside the region where normal imd measurement method can identify imd.
From two different multitone measurements, the one amirm used showed no weird modulations, but when used different spacing between frequencies and changed the slop of the frequency energy, it excites the modulations. It's there from the max volume to drowned in the noise floor.
With two tone 20hz/30hz + 1000/1111hz (about this range) shows the modulation can be quite high and lands in audible frequency range instead of 30hz+7k which pushes the modulation to higher frequency causing people to mislead to think it's not audible. It can be very audible, i hear big difference in imaging, and "sour" sounding similar to mp3 etc. If excited with a high amplitude low frequency and a low amplitude 1khz, the modulation can go higher than -80db reference to 1khz. Since music is very dynamic, instantaneous modulation can be very high. Multitone is averaged response which averaged out the distortions. So it can be audible even when the graph shows it's lower than audible threshold.