[cx01]

You do not copy data to all servers; in practice you may choose a replication degree of 3, so all data is written to 3 servers. In chain replication the first server in the chain will decide on the order of writes (i.e. he will serialize the writes), which is very fast. In a P2P system you'll need to do some kind of distributed consensus for every write, which (when conflicts happen) will be much slower than having one node serialize the writes.

Having multiple masters will most likely add overhead for conflict resolution and should be avoided, but it's really orthogonal to the whole CAP stuff. If you allow writes to multiple masters without conflict resolution at write-time, then you will need to do the conflict-resolution at read-time, which is going to be much more complicated, because your read may yield multiple versions.

[kennu]

That was pretty much my original point about giving up consistency to achieve (write) scalability. E.g. CouchDB is based on that principle: You can write to all masters in parallel and the system provides a mechanism for resolving the conflicts afterwards.