The documents should be serialized into BSON and output one after another. As BSON is length-prefixed, there's no utility in designing a serialization that requires an explicit delimiter between documents.