That's the one bit where Crinacle's rig differs a bit from hammerhead style fixtures. Unless I'm mistaken in the ear canal gain region (1-4kHz or so) HPs still operate under "pressure conditions" (I don't know the right terminology, ie basically SPL is the same at any point in space inside the front volume) - that's what that article suggests for example :
https://www.aes.org/e-lib/browse.cfm?elib=20605
So anything that affects the front volume's, well, volume, may affect the response, and perhaps in the case of the APM disproportionally so in the ear canal gain region. And the APM's headband to cup attachment + spring loaded earcup may add quite a bit of difficulty to the equation. Perhaps the ways pads compress as well : on a rig with a flat plate around the pinna it's probably quite a lot more evenly compressed than on a real human's head (particularly since the APM's pivot + spring cannot by design
ensure an even pressure around the ear for most people).
Above 4-5kHz no idea what's going on.
Amir's results don't seem to diverge any more from the average of the APM's results on a GRAS rig than the other individual results. It's a generalised crapshoot past 1kHz or so.
As I said I've measured four APMs, that's eight earcups in total (since they're nearly perfectly front to back symmetrical, unlike other headphones, it's somewhat valid to reverse them and measure them on the same ear), and the response didn't vary that much (if taking into account the typical seatings to seatings variation I see in the ear canal region on my head with the APM there may have been no more than around 1dB of sample variation in that range, with one cup being an outlier and the others even more tightly grouped). Cf graph for four earcups earlier.