Exactly that.
To me the emphasis on the presence region feels like they are trying to force the frontal localization. IMO this sounds rather fake because it limits the preceived width of the soundstage and makes the sound very hard and tinny. It adds fatigue.
Can't disagree about the fatigue happening. VERY recording dependent, though.
Let's think it through. If we:
- assume the curve is was taken from speakers angled some degrees (I don't know exactly how much it was) from the listener,
- the resultant ear behaviour in itself is variable due to angle,
then it'd be natural to assume, that the curve derived that way would remind us of such a speaker setup. Whether it's good or not - different topic.
I think it's possible that we're being hit here by the "pan law", which makes the center sound a different volume on speakers and on headphones. I'd try to explain what it more, but I don't get the thing that well to explain it properly, yet I feel that's the proper way to think through. It might result in phantom center being "smaller" on some recordings. It might not just be Harman sounding wrong; it might be that speaker HRTF w/o cross-ear interaction on headphones just doesn't sound "true" and will never be truly universal, working on many cases but not on all. Spatial perception isn't the matter of the Harman "preference experiment", anyways.
Also, if you think how binaural recordings are produced, it makes some sense to think that way: the crossfeed is taken "into the recording". Harman's tonal character could actually be neutral, yet the presentation as a whole wouldn't be and would probably ask for driving too loud, since stereo doesn't take the crossfeed. In the end, you'd end up with either harsh/tinny recordings (because of distant centre due to panning "for speakers" with a different pan law) and the ones suffering in a variable amount from the combination of these, plus the recordings done "wrong" for speakers, but sounding fine on Harmans (due to not sounding "too distant" when on proper loudness). Maaaaaaaaan!
I also wonder whether Harman works better on open vs closed cans. If the problem was crossfeed-related (not just crossfeed, the whole room interaction), I believe open-backed designs produce some amount of crossfeed and that could benefit the tinny/narrow character. See - Sundaras are open-backed cans with nearly-Harman (sparing the bass) FR and they're pretty popular & not mentioned as harsh, not so much, at least. I've seen numerous people bothered with K371s being problematic.
Not all but most of the headphones I own have this issue when EQ'd to Harman (following Oratory's settings). It's not perfect. Still a good starting point as already mentioned.
Well, oratory mentions that he listens for distortion happening - but due to production variance you might get a pair with imbalanced channels, where one driver starts resonating and the one isn't yet. Going further...
... the whole "let's EQ everything to Harman" affair shows up the inherent problems of statistical stuff - how averaging things mocks up the details, sometimes absolutely ruining the sense. Some AutoEQ measurements show absolute nonsense due to averaged measurements, when you dig into the details. Crinacle provides three original measurements + L/R separated graph (obviously not accounting for the variance), o1990 doesn't give out neither original measurements nor the L/R separated ones. Hm.
For me, the bass should be somewhere between the flat in-room response and the average preference curve. Then a little bit of attenuation in the presence area. Works quite well for me.
Wonder whether the preference score is tied to how long you have to listen to stuff. Maybe endurance-testing of curves would show us more reliable information.
It'd be necessary to provide example material for people against Harman to show them that it works sometimes and some stuff to show that oftentimes it just doesn't work for the rest. We'd probably all gain a lot of perspective from that.