I still have some questions about the above. But a couple links in DMS’s video led me to the video and article below which explain some of the rationale behind compensation curves like JM-1, which are based on a population average.
My understanding is that the 5128 ear canal and coupler are based on an average of a number of humans. So I could see some justification for using an average human DFHRTF for measurements of earphones inserted inside the 5128 canal.
If the goal is to see how both earphones and headphones behave on an average human though, then a better solution is to equip the 5128 rig with a different head and ears that measure closer to that.
DF is not the preferred response of most listeners though. Most listeners prefer a response close to the in-ear response of neutral speakers in a semi-reflective room. So that is really the reference point we should be going for at this point imho.