See my answer above - in cases where the participants are all known you can use BFT algorithms to prevent attacks so long as less than a third of the replicas are malicious.