The order returned from the Index Scan is not the ordering requested by the user, so there would still have to be a full (or topk) Sort over the dataset returned from the index scan, which could negate the gains you get from using an Index Scan; PostgreSQL itself does not produce merge join plans that merge a spread of index scans to get suffix-ordered data out of an index.