I don't read this article, because too long and complicated for me. But ultrasonic perception of your idea (some nonlinear effect in the hearing chain) seems good point. Not only ultrasonic frequency but also am radio frequency can be heared by non-electrical (which means without electronic radio reciever) mixing down (product detecting) in the very close area of AM radio transmitter antenna.
Just for information:
I tried to prove my nonlinear downmix theory by generating 2 testsignals, one with high frequency content, the other without.
1st Testsignal without high frequency content is a quality 24/96 broadband white noise (low level, unclipped, constant spectrum up to cutoff), LP-filtered at 18KHz w 48dB/oct.
2nd Testsignal with high frequency content is the above 18KHz LP-filtered signal plus 22KHz HP-filtered (48dB/oct) white noise from the a. m. source. This avoids any level overshoot in the cutover region and maintains identical content in the audible frequency range. The level is of course higher, but far away from clipping.
Now here comes the result:
I was not able to detect any difference, under several loudness conditions. So no nonlinear downmix detectable in this experiment.
Also tried running the LP-filtered signal on left and only HP-filtered content on the right speaker at once and move right next to the right speaker (more HF-level). Also no difference audible, while switching the right channel on and off.
As described, I could hear a difference in that a. m. downsampled album under the same conditions, so I guess I need to attack the issue differently...