PostgreSQL pain points

It's a real problem, but the solution isn't obvious and there is no way for kernel developers to reproduce the problem locally. Tackling a problem like that wouldn't fill me with enthusiasm either.

Seems to me that the priority for the postgresql team is to provide an easy way to trigger these problems - ie as suggested, a small app that generates the access patterns that postgresql does. I bet that if such an app were part of one of the regular kernel test-suites then the problem would be fixed and never come back.