I've reviewed logs for btrfs related to orphan inode handling but haven't been able to see an obvious candidate causing the problem. Looking at the various patches between 4.4 and 4.13 handling corner cases where btrfs send/receive may fail makes me wonder how reliable this really is, though.