jesper@krogh.cc writes:
> Secondly I could bump the default cost of ts_match_vq/ts_match_qv a
> bit up, since the cost of doing that computation is probably not as
> cheap as a ordinary boolean function.
Actually, you could try bumping their costs up by more than "a bit".
It's a tad unfair to blame the toast access costs on the functions,
but it might get close enough to what you need. If memory serves,
we don't charge those costs against index scans.
regards, tom lane