I'd love to have your ideas on the matter. But I invite you to re-measure your QC25 on your own head with varying amounts of pad compression.
Difficult to be conclusive without more extensive repeated testing, but my guess is that one of the variables that varies when compressing pads is also present as a variable across individuals. Perhaps volume of air in the front volume, which could vary between individuals because of varying geometry around the pinna, something ear simulators can't test particularly well for (and none of which are anatomically accurate around the pinna to start with).
Some headphones, and in particular some active headphones, and even more so in particular some Bose/Sony/etc. headphones, are
extremely sensitive in some part of the spectrum above the range where their feedback mechanism operates and below a few kHz or so to pad compression (while their feedback mechanism makes them, on the other hand, extremely insensitive to pad compression or small leaks in the range where it operates).
Some ideas thrown in these posts :
As I said, design criteria is not proof of accomplishment. Independent testing and research needs to be performed to ascertain that. Are you suggesting that there are issues with B&K's underpinning research? It seems to agree well with a randomly grabbed paper from an unrelated source, and, of...
audiosciencereview.com
Forgot to give credit to @Sean Olive in his tweet here. I believe this is an actual production sample - it slightly differs how it measures from the production validation sample posted earlier. What @Sean Olive thinks about Brent Butterworth´s measurements at Soundstage Solo...
audiosciencereview.com
These headphones also tend to show the highest amount of variation between ear simulators in that "above feedback range" - "below a few kHz" range.
Ex : QC35 in Sean Olive's presentation for the HBK conference :
Rtings also tested the 5128 against their current ear simulator and SamV released some files in the wild.
I've been as a result very wary of over-interpreting
some headphones' ear simulator results without additional information on their behaviour under several variables. For headphones such as the Bose 700, QC45 or Airpods Max, I've found ear simulator results to be a rather poor predictor of their actual on-head behaviour
for me.