The existing documentation in Linux Memory Overcommit seemed to
assume that PostgreSQL itself could never be the problem, or at
least it didn't tell you what to do about it.

Per discussion with Craig Ringer and Kevin Grittner.