These are excellent questions gents, and for a while I have been looking for a copy of the
original paper published by the AES in 2014 and the follow-up study in 2019. From what I know, they collected data from 238 participants of varying age, gender, and had a mixture of trained and untrained listeners in four countries. They were given a headphone with an approximation of the Harman curve
and allowed to freely adjust bass and treble to their preference.
Imo this still raises the question "did they try using different corner frequencies for the bass & treble adjustments or not, and if not, how did they pick those frequencies?"
Considering that this is just a hobbyist question I'm not going to purchase papers from AES to see if they mention this. It is not mentioned in
this summary pdf linked above. However what
is shown in the pdf is that most headphones they tested, including those rated as "excellent",
did not have the 250 Hz dip and instead seemed to roughly copy what seems to be my personal preference. Though the deviation seems to be centered around 200-230 Hz rather than 250 Hz, but that may be just my own imprecision, I did compensate for the dip only by ear and secondhand measurements.
You can see it here, the average frequency response (in blue) is relatively flat over the midbass region, so it deviates from the Harman target (in green):
Pretty interesting.