Just show that distortion is significant/audible beyond certain level, then do not measure/publish the frequency response beyond that level. At that point the whole discussion about "multitone intermodulation torture test " becomes utterly irrelevant to the frequency response . It is that simple.