Perhaps I'm just splitting hairs here, but would there really need to be as many as literally one per pipe - eg not one per two pipes or one per four pipes, ... ? Still an awful lot of convolution reverbs to run at once, of course. But I was just wondering how coarsely spaced it would be possible to make the "grid" of IRs before it was possible to hear the difference from a one-per-pipe grid. I suppose we won't know til someone actually tries it?
Very slight differences in position can give significant differences in the pattern of early reflections (in an impulse response or wet sample) and the frequency response heard. For example the sound from one pipe might pass between other pipes on its direct path to the listener, whereas the sound from its neighbour might be reflected or refracted off another pipe or part of the case before it reaches the listener.
The subconscious brain is very good at detecting those slight differences in order to build up its spatial sound impression. E.g. if you close your eyes and play each pipe in a rank on a real pipe organ you can often tell its position with remarkable accuracy just from its sound as you play and release it, and relative to its neighbours.
That ability of the brain is largely responsible for allowing us to identify and hear each pipe separately within a real pipe organ, or within a wet sample set (or potentially a dry one with separate convolutions for every pipe), even if many pipes are sounding at once, giving perceived clarity.
If the same impulse response/reverb is applied to several dry pipes (or even all the pipes) then the brain is no longer so easily able to hear them as separate pipes/sound sources, and the result will lose perceived clarity.
It's exactly the same as playing a (dry) digital organ in a reverberant building - the more speakers you use, the better the spatial impression and clarity. Too few speakers and it can sound artificial and muddy. Each speaker equates to a separate convolver instance and impulse reponse when adding an acoustic virtually.
When I helped Brett with recording some of the original demos for the (dry) Virginia WuliTzer sample set using convolution and simple room modelling software to synthesize the IRs for a virtual theatre, we found that about 32 separate true stereo (2->4->2) IRs and convolution instances (64 standard stereo convolutions) gave a level of clarity and spatial impression for that 8-rank organ that began to sound fairly convincing to us. So that would equate to using about 32 speakers for a dry digital organ in a real reverberant room (four per rank on average).
The more speakers/IRs/convolvers, the better the spatial impression and perceived clarity. One for every two pipes would almost certainly sound extremely good indeed, but I suspect you would still be able to tell the difference in a blind test between that and one per pipe (which a wet set effectively achieves).
(There are other possible compromises, for example you could make a 'semi-dry' sample set that had only the early reflections within the samples, and then use a reverb unit or IR just to add the background wash of late reflections, which are less important to the brain in terms of spatial cues. That approach would have some of the advantages and some of the disadvantages of both wet and dry.)