This happens because the shard will continue to process the remainder of the batch even if it encountered a version mismatch. This in turn will generate a response with multiple stale errors. And when the mongos processes the response it will trigger a dassert that checks that only 1 shard can have a stale error.