Think of an industry e.g. like banking, where the different participants know who each other are but don't fully trust each other. Presuming each participant controls one replica in the system, then using BFT algorithms you can ensure that so long as less than a third of the participants are malicious it is not possible to perform such an attack.