This is a tricky one, because our current productivity scoring calculation would favor an index scan. However, an index scan is probably a better choice than a distinct scan for keys that have few/no duplicates, because we don't seek for every key (we just scan). This will require some benchmarking work to confirm as well.