> It really has very little impact. It only affects index scans, and
> even then only if effective_cache_size is less than the size of the
> table.
>
> Essentially, when this kicks in, it models the effect that if you are
> index scanning a table much larger than the size of your cache, you
> might have to reread some blocks that you previously read in during
> *that same index scan*.
Ok, thanks for clearing that up for me. Still, I think the doc could be
improved on this point (sorry to be a bit obsessed with that, but I'm one of
the french translators, so I like the doc to be perfect :) )