I thought this sort of thing was pretty much impossible due to latency issues.
There are other solutions, however. Just have a look at any FPS (first person shooter) game. These types of games are extremely latency sensitive, because a win/loss in any direct engagement comes down to milliseconds. There are entire areas of study dedicated to working around latency in games. I wouldn't be surprised if some of the techniques could be applied to this type of real-time collaboration.
Musical tracking has to take special measures to keep the latency low enough even when its all local; see the need for ASIO, jackd, etc.