I haven't checked Bitmessage's protocals, but that it is not nessasary to identify the intended recipient. If you publish the ciphertext with no meta-data, then everyone can attempt to decrypt it. Only the intended recipient would succeed, but an ease dropper would not be able to identify who successfully decrypted the message.
This raises an obvious scalability question, as you need to attempt to decrypt every message that is used in the network (in addition to simply downloading it which you have to do anyway). There are probably ways to mitigate this overhead without compromising security.
If you read the wikipedia entry is says that the destination id is required. I assume these must be created somehow and communicated to the sender before being able to receive anything.
So long as this is occurring, the way the scheme works, unless you download every message, it could be determined which id belongs to whom. Once you determine that it is easy to track originating packets to that destination since it is not an onion system.
Additionally, without side channel communication ( in person or posting that address somewhere publicly ) it is impossible to anonymously communicate to an intended recipient.
