This is why I always had the impression that sampled reverb (i.e. convolution reverb) interacts better with sampled instruments than synthetic reverb (like the Lexicons).
William's impression that IRs from those synthetic reverbs never sound as the original has a simple reason. Most reverb algorithms, and especially those by Lexcion rely a lot on the modulation of the reverb tail - something static convolution can't mimic by definition. A real room is more or less a linear system, something static convolution is able to mimic very well.
Mathis' "instrument idea" is exactly the reason why we invented the MIR, BTW. The underlying concept is all about multi-sampling a room, just as we did with all those instruments.
/Dietz - Vienna Symphonic Library