IMO soundstage is more of a 'brain' thing.
That is, how well your brain can recreate a 'soundstage' from the signals coming in from just 2 point sources close to the ears without the assitance of wall reflections.
I am quite capable of hearing 3D soundstages from a good speaker system though.
Amplitude and phase relation between the 2 drivers as well as phase relation of the driver itself and your hearing/brain will determine how 'good' the soundstage is.
Personally I never heard a soundstage, not even with binaural recordings, but just because my brain cannot recreate a 3 dimensional image doesn't mean others cannot.
Those that are fortunate ? enough to create a 3 dimensional image, instead of a simple L-R and inbetween image, might enjoy headphones more if that aspect is good as well.

Needless to say if a HP measures/performs well in the earlier addressed areas (amplitude, phase relation and perhaps decay) it will be easier to recreate a soundstage if your brain can do so.

detail retrieval has to do with how well/accurate the diaphragm can follow low level (usually high frequency) small signals that occur simultaniously with larger excursions.
The better the FR in the top and the cleaner the CSD and the lower the distortion the better the detail retrieval.
Also higher output in the FR above say 6kHz COMBINED with a good treble extension without too many peaks and dips will also greatly benefit detail retrieval.
Muddy bass can overshadow details as well.

The way I see it, "Detail retrieval" of a transducer can be more simply regarded as the amount of source information that it can pass through to the ear/brain hearing apparatus.There need to be two(that I can think of) requirements iin a transucer to help it perform this function better:

1) A flat frequency response : Its well known that any bump in FR is bound to 'mask' the information in nearby frequencies, a recess would conversely be masked by the surrounding frequencies. So, your best bet is a perfectly neutral headphone.

2) Low distortion : Since any distortion is basically information added to the signal which wasn't originally there so excess of it is bound to harm the detail retrieval.Like Jerg said, big thin diaphragms have a better chance of achieving that.

I'd like to add a very important 3) to that being impulse response. There are several reasons for this, one you have already covered partially being if the diaphragm can not start and stop movement fast enough, resonance/distortion or artificially added sounds will color the sound and lower the perceived clearness of the sound, and thus the notability of details.
But also how quickly and how strongly the diaphragm responds to a signal in comparison to that following resonance. the steeper this graph is, the more accurate the reproduction will be, and the higher the initial impulse peak is in comparison to the following, the clearer the details will stand out, and thus be perceived. Which is to some extent what some manufacturers are trying to amplify, by making the high frequency response of the driver stronger than the rest, because those peaks will create a strong contrast, and thus a higher perceived amount of detail.

As for soundstage, this as mentioned is a product of our brain, but there are ways to simulate/make it occur, the easies of which are to use the outer ear that normally creates the latencies that tell us where a given sound is coming from, simply by making a supra-aural design that fires the sound at as large a part of the ear as possible, at as natural an angle as possible. This is the reason why fx. HD800 and K1000 are some of the masters of soundstage, because their massive cupsizes or lack there of, and positioning of drivers, lets the brain process the sound they produce the exact same way it would if you were actually facing the source for the sound/the band.
What this also means is that to some extent a headphones ability to reproduce the feeling for 'perceived space' is a matter of physics, which cant be seen on a graph. The bit that can though is that for this to work, the driver must also be able to deliver these latencies in the sounds that are normally caused by the reflections of a room and/or the distances of different sound sources, which brings us back to the first topic: the headphones ability to reproduce details accurately, which can be measured.

That my take on it anyway... 

I feel the treble accentuated 'detail' is the fake sort.  Bright phones like many Beyers do this.  Other phones won't lift the treble but just roll off the bass like the SA5000 and Shure 1840.  Though the SA5000 does do real detail retrieval quite well too compared to other bass light fakery.  What amazes about the HD800 is detail all across the spectrum.  Yes it's bright, but the mids to upper bass have a lot of information.  The Audezes and HE400 have amazing bass and sub bass detail retrieval.

Yes, this is also why I like to describe it as 'perceived detail' and not actual detail retrieval - it is all about contrast. Actual detail retrieval is much harder but yes, though the HD800 is partially guilty of this though not extensively, as you said, it is very very good across the board none the less, which is why it is currently king in my book (havent heard the SR009 yet though, but that'll soon change...).

I'd like to add a very important 3) to that being impulse response.

I did consider the impulse response but IMO while it is probably a good measure of a driver's technical ability , it doesn't always translate into actual detail retrieval.Their are drivers which have excellent impulse response but are severely limited in reproducing the entire frequency range. Also, if this driver has significant distortion, peaks, ringing etc. then a measured 'fast' impulse response won't amount to much as far as detail retrieval is concerned.
What I mean to say is that even a 'fast' driver has to reproduce the entire range without significant bump/recesses & distortion to be able to extract the information buried in the source.