Robert Haas <robertmhaas@gmail.com> writes:
> If you left seq_page_cost (which isn't mentioned here) at the default
> value but reduced random_page_cost to 0.1, then you have
> random_page_cost < seq_page_cost. That's probably Bad.
... well, it's certainly going to push the planner to believe indexscans
are cheaper than sorts no matter what.
The previously noted rowcount estimation problem might be a bigger issue
in this particular case, but I agree this is a Bad Idea.
regards, tom lane