"Matt Clark" <matt(at)ymogen(dot)net> writes:
> I've noticed that the cost estimates for a lot of my queries are
> consistently far to high.
You seem to be under a misapprehension. The cost estimates are not
in units of milliseconds, they are on an arbitrary scale with 1.0
defined as one disk fetch.
LIMIT throws another monkey wrench into the mix: the estimates for the
plan nodes underneath the limit are done as if the plan were to be
executed to completion, which of course it won't be.
regards, tom lane