Personally,I think I'm favoring reaching an identical view of the room between participants before starting the conversation because:
1. It is explicitly mentioned as the goal of entity authentication in the first mpOTR paper and we are mandated by the contract to have the same security properties bar forgibility and deniability:
"The entity authentication goal for mpOTR is to provide a consistent view of chatroom participants: each chat participant should have the same view of the chatroom membership" (GUVC09)
2. Consistancy is much simpler to check when we know everybody's view was consistant before they start talking.
3. It is more modular approach: first check the view consistancy and when done with it don't worry about it during the chat. This makes the whole implementation simpler.