The resulting sound will be glitchy as in notes that were not played will briefly be heard--similar to how a character position might glitch in an MMO. Anything without predictive sound broadcast wouldn't have the timing to be fun to play together is basically my conclusion. I would love to see this, as glitchy as it might be would still be awesome. Maybe something that uses both TCP and UDP might help. UDP to get early input changes, and TCP to always eventually catch changes.
Yes, there is latency over the internet, but IMO you can get used to it.