I agree with your rationale, and these were my initial thoughts as well. But I suggest making a test map and playtesting it before locking it down. Music and UI sounds are no-brainer, but I wouldn't be surprised if e.g. excluding all player grunts from EFX actually sounded better than having some sounds with, and some without EFX.
Sometimes consistency beats logic, even if it's counter-intuitive
I don't agree with you. Testing on a single map is not a good solution.
EFX is not like gamma correction.
For gamma correction, you do it once either in your texture, or in your engine. In any case, you only have to ensure that it is done exactly once, but it does not matter where.
EFX is like lighting. You can bake lighting effects into texture (e.g. lightmaps or ambient occlusion), but then you loose the ability to change light sources around it. As a result, a texture with baked lightmaps can only be used in a single place on a single map. If we bake reverb into a sound, it looses the ability to reflect the environment around it. It would sound well in one place, but weird on another place. The fact that a sound with baked reverb sounds well without EFX when you are in one particular area (with some EFX effect) does not mean that it would sound appropriate without EFX in all other areas (with different EFX effects).
Of course, this also depends on how different are the EFX effects you can create. If all of them are mostly equivalent to "apply reverb of strength X", then maybe it does not really matter.
So the only "correct" solution is to decide which sounds should logically be affected by surrounding environment, and which should not.
If any sound already has reverb and must be affected (or has no reverb and must not be affected), the only "correct" solution is to fix the sound sample.
The problem is that this "correct" solution may be too hard to achieve, so we need some ad hoc solution for 2.06.