A very important thing to remember is that a convolver places an audio channel fed into it at a single point in the virtual acoustic. So if you use a dry sample set with stereo output and feed that into a convolver then you'll hear the sound come from only two points in the virtual acoustic (with the L and R channels corresponding to the positions of the two microphones used when recording the impulse response, or the virtual microphones if it was synthesized).
From a theoretical point of view, to get the same spatial impression that a wet sample set can give (which allows the brain to separate the pipe sounds more clearly because of their slight differences in acoustic, and gives an increase in perceived clarity as a result), i.e. with each pipe emanating from a different point in the virtual acoustic, you would need one convolver instance and a separate impulse response, per pipe.
Current computers can't handle that many convolutions for organs of any realistic size, so there is inevitably always a compromise in terms of clarify and spatial impression by using convolution, when compared to wet sampling. The situation is basically the same as using dry sample sets in a real acoustic (the more amplification channels, corresponding to the more simultaneous convolvers, the more realistic the virtual acoustic and the higher the perceived clarity).
However, of course wet sampling has other compromises (tremulants are less realistic, no ability to adjust the reverb/acoustic, more memory required, etc.). There are lots of previous forum posts discussing the pros and cons at length, e.g.:
http://forum.hauptwerk.com/viewtopic.php?t=2293
Best regards,
Martin.