I don't think your attack violates the security model of this protocol. One of the proofs in the paper is that if you have a network of good and bad nodes the bad nodes can use tricks like you posted to discover whether a good or bad node sent the message. I think that's all they can discover: did an attacker send this or not? But then again they already knew that, didn't they? So I think your attack is only interesting for 3 nodes.