That's exactly why I've started to measure headphones on my own head with various techniques, for the "last mile" in EQ that you won't get from dummy head / test rig measurements. If I equalise HPs according to someone else's measurements and profiles to a target, they'll still diverge quite significantly when measured on my head (albeit less so if they diverged a lot before).
Not that dummy head measurements are useless, far from it, given that most HPs are still grossly deficient and would benefit from third party EQ profiles (although a lot are rubbish enough that they have un-equable peaks / dips and can't be saved). That's a great service to the community what Oratory, Crinacle, Amir, Rtings, Resolve, etc. are doing.
Otherwise it's evident that FR can't be all in terms of sound localisation with headphones since you're necessarily missing the other cues we use to localise sounds in space. For that we need surround sound simulation systems for headphones, perhaps tailored to one's own anatomy.
Do you use crossfeed algorithms ? Personally I don't get anything that resembles speakers without some degree of crossfeed, and the latter's algorithms were utterly unconvincing until I started to equalise my headphones in a more fine-grained manner (and they're still not that great. I'm not sure that they will ever be until the algorithm can be modulated according to my own morphology, at least to some point).