[...snipped]
Just an idea (non tested!) on compression at mastering time: assuming a
mapping from voices to bandwidth may sometime be not relevant: that way
would cause compression issues on stereophonic takes (piano, cymbals
overheads, any ambiance pair). May be switching from the R+L master to
an M+S working copy (non-destructive) would help to isolate accurately
instruments and apply compression on the M track only ? This is an open
question. (*) M = (Left + Right)/2 ; S = (Left - Right)/2
- ben