Ever since Etherpad/Google Wave/Google Docs, though, OT has been consistently mangled into a hub-and-spoke design, where there's a server who keeps a canonical state, and the OT tie-breaking algorithm is always decided in the server's favor.
Used in this fashion, it's not much different (actually a bit higher-overhead than) just having the server arbitrarily accept and temporally-order input operations in a way it likes, and then send back a single overwriting transformation to all clients to move them from their last server-known commits to the new server-canonical commit. (In other words, what any multiplayer game's logic server does.)
[1] One major difference is that OT "commits" are referred to by their vector-clock, instead of by their SHA. Cryptographic hashes are nice, and would give OT even cooler properties if it could use them, but they're a bit too slow for something that gets regenerated with every character typed. CRCs might be fast enough, but in an actually-distributed peer-to-peer system you'll get real collisions fast.