If you're recording something fragile, exposed and LIVE -- a string quartet plus guitar plus vocalist, maybe -- then I would opt for 96k.
Unless, however, you're working with live recordings, there is only the tiniest, potentially inaudible benefit of using a sampling rate higher than the original samples (which typically are 44.1 or 48, though of course some are 96).
I've heard arguments about benefits of 96k for reverb and some other FX but I wonder how audible that really is? Certainly if you're writing for TV / movies / games where the music is not the only thing you hear, I doubt any but the most bat-eared will be able to detect such marginal differences.
And if there is dialogue / background / SFX on top of music, probably not even the bats.