I think the win for that kind of scenario is smaller than you make it out to be. I agree on memory usage, but the concurrency argument seems largely faulty beca

I think the win for that kind of scenario is smaller than you make it out to be. I agree on memory usage, but the concurrency argument seems largely faulty because you at the same time increase page-level contention on a shared resource (postgres' buffer cache). I personally want a lazily filled hash-join / something like a index nested loop join that has a fixed size cache for both positive and negative lookups.