Has anyone done any testing with decently large numbers of hosts and replication? I would like to know some practical limits as to the number of replicated hosts that a master can handle.