Well, that would partly depend on the consensus pool as they may have their own blacklisting mechanisms, but broadly every message in the system is signed by the client making the request, and the consensus node passing it on. Every signature is published, so consensus nodes passing on messages that were incorrectly signed can be detected very easily. There are other attacks (not forwarding any messages is a simple one) that can be used, so pools will have to implement their own policies on timeouts/uptime etc.