The author dismisses the RULE based approach. I'd love to see it compared, as it was the only thing that I could get to complete for a multi-GB table merge (I can't recall what other approaches I tried). http://stackoverflow.com/a/6176044/6691
It would be a boon if upsert was supported as a declarative primitive in psql so that the query planner can do it's magic. Pity it didn't make it in 9.4.
Author here. I dismissed RULEs and TRIGGERs since they don't work correctly under concurrency (and most RULE-based approaches are broken even without concurrency). I felt it wasn't a fair comparison.
However, I guess it would be nice to know how much of a gain would be possible if one was willing to forfeit correct behaviour under concurrency. But that would probably require more work than I'm willing to put in, since the answer doesn't interest me that much personally at this moment.
It would be a boon if upsert was supported as a declarative primitive in psql so that the query planner can do it's magic. Pity it didn't make it in 9.4.