Resonance just sounds bad.  If it doesn't sound bad to you, then it's not "bad".  some people , esp on HF seem to enjoy resonance in certain spots and that's certainly not a "bad" thing if it's what you enjoy.  You can still use the CSD plots to learn more about what kinds of resonance sounds good to you.  But as Marv tried to point out in an informal study he conducted here, there is a strong correlation among experienced listeners between headphones sounding good and having clean response. 

As for the closeness of the ear, that's a very common argument, but it's not entirely logical IMO.  The point is for an audio device to measure flat, and sound flat by the human ear.  If the ear is closer to the driver, that driver needs to measure flat closer to the driver.  Measuring flat at any other distance is irrelevant because the goal and reference is the human experience.  Headphones don't need "shelved response to sound neutral because of the proximity to the ear" (the common argument).  Why would a headphone be measured and judged in different way/environment from it's intended use, and then need to be altered for it's intended use?  Shelved from what?  Shelved from some irrelevant measurement standard? 

Would be like saying that a car performs really well for space travel, but since it won't be used in zero gravity environments, it needs to be modified.  You judge things by their intended use.  It doesn't really matter how well a car performs in zero gravity, because it's sole purpose is to be used on the ground. 

These are just my thoughts, from my experience.       Maybe someone else here has a different opinion.

Wouldn't listening to a pair of speakers at 6 meters and a pair of headphones at 6 cm with the same frequency curve sound very different?

Or did I drink too much this afternoon?

You are over thinking it. Speakers measured at 6 meters with the same FR (taking into account the influence of early room reflections) and headphones at 2 centimeters with the same effective FR (taking into account the influence of cup refraction and reflections) should sound the same. This is why I compensate my headphone measurements (calibrated by ear with near-field monitors which measure flat and with other reference headphones) to represent what we would hear with speakers in an ideal setting.

The neutrality thing is simply a reference. I actually flip a switch on my crossover to have a slight roll-off for actual listening in my usual listening position. The HE90 FR (imagine it averaged without the bumps from 1k onward) is probably more representative of my own personal ideal FR graph vs. a flat line from 20Hz to 20kHz.

Many headphone hobbyists, prefer a bass hump because headphones lack the tactile feel of speakers. For me, I find this cure worse than the disease and would much rather run headphones with a subwoofer with a slow roll off starting at 35Hz (depending upon headphone) if I wanted that tactile feel.

Finally you may want to read this before I delete it in a few more days: http://www.changstar.com/index.php/topic,398.msg6630.html#msg6630

Purrin should correct me if I'm wrong, but seeing the compensation curve would only tell you about the measurement rig, and the ways it differs from the human ear.  The compensation curve is like zero-ing out a scale before weighing something.  In other words, uncompensated curves AFAIK are just as much a measurement of the rig as they are a measurement of the headphone.  So it's not really that meaningful. 

I believe you're thinking there is some kind of pure response that is neutral, and which is not related to what we hear.  But what I'm arguing is that the way we hear it is the only thing that matters. 

Of course Purrin's measurements will always be slightly different from Tyll's measurements. Different headphones, different measurement techniques, different compensation curve used.

Tyll uses HEAD Acoustic's own ID curve for compensation. Some of us thinks that this curve might be causing artifacts in his graphs.

Uhm, to make open headphones, then compensated, sound exactly like speakers, you'd still have to introduce inter-aural time delay and level difference simulation. Simple such simulation would be a crossfeed circuit.
There's also head related frequency response change (HRTF) affecting the high end > 2kHz or so - those are emulated by some headphones too, but differ notably from person to person.

None of this involves any large shelf cuts or boosts - lack of those is what's known as neutrality. Also all of the above is negatively affected by distortion or ringing which is not present in normal hearing at all.