One StackExchange answer mentioned bitmessages can be traced, another said it is fully anonymous. The second link seems to cite Wikipedia but a user points out that this anonymous concept doesn't exist in the whitepaper.
My concept is that you could have a tremendous number of bitmessage clients that track the IP address of every client that sends you a message. You keep a log of the IP address and the messages it sent. It seems like you could make a bet of fairly high accuracy whether a client is originating new messages when it's repeatedly the first source of a message you have not seen before.
Is there some aspect of the bitmessage protocol that makes this impossible?