Ok. That sort of is what's being done, but not so finely grained. Real-time processing can be a bit of a cpu hog, but some of the simpler effects can be done easily (either cpu based or through basic intergrated sound hardware).
Take DoomRL's stereo-panning. Light on the cpu, but is actually a part of the game mechanics. I don't think kyzrati (although I don't know) wants to make it too complicated, but I think X@com had stereo-panning and volume adjustment for distance, with different samples played for different environmental conditions. It can get very complicated, but it doesn't have to be.
Rather than a gun being made up of different "bits" of sound, they'll just have their own unique sound. Ricochets are probably "weapon type" based. So sort of like what you're saying, but in bigger chunks. Yes, what you're describing can be done, but it tends to sound worse than a large pre-built library of unique samples. Pitch shifting, volume, delay, stereo panning, etc can still be done to them, but they sound better "as a whole", rather than as a composite of pieces.
ps. It's a good idea though, depending on the game. In a "real world" based game, slightly futuristic, where you develop new weapons and technology based on current technology, it could actually be very handy. A new 5.56mm rifle got developed by your civilization? The engine could put together a fairly unique sound scheme for the weapon, just for you