I've used PostgreSQL for 15 years and missed upsert/MERGE for all of them, but my understanding from previous research is that MySQL's is actually unreliable in the sense that it can violate ACID in some circumstances. This is believable to me, MySQL's performance lead on whole table count(*) is because it doesn't actually count the currently visible rows for large tables. Complete violation of sql's count semantics, but most people don't know or care.
I'm digging up some references now.