Just to put my own preferences in perspective, my company developed and patented a special system for electronically altering room acoustics. The main person behind this system engineered almost half of Decca's classical recordings, and that is one of our main target audiences. Not sound reinforcement, as that replaces the point source. But it's WAY more advanced than convolution (which is a part of the process nonetheless).
Dialing in the right amount so that it sounds natural, as though it is part of the room vs. altering the instrument's sense of where it is, is the hardest part, so we have to have highly trained operators who deal with that. And LOTS of microphones are involved!
Anyway, being semi-involved in that, watching Vienna MIR, WIVI, and other approaches to spatial aspects of acoustics, has made me more sensitive to baked-in large hall sounds in sample libraries, but also aware of what can be added vs. enhanced. So yes, 100% dry/anechoic isn't good either -- this is why the Sample Modeling stuff, other than for the lower-frequency instruments, can be hard to work convincingly into an orchestral perspective vs. using it for jazz, pop, reggae/ska, etc.