Reading the code, it appears to me that both branches of _mongoc_write_command_execute (_mongoc_write_opmsg and _mongoc_write_opquery) skip a document that exceeds maxBsonObjectSize and then continue sending the rest of the documents in the batch. They should conform to PyMongo and (I think) other drivers: a single oversized document aborts the bulk operation.

Doesn't matter if the bulk op is ordered or unordered.