I haven't used it but have looked at it. It caches a list of keys and sets up all the "next" and "prev" links for you. Then inside the loop you select to get the right rows. The theory is that this second select is cheaper than your typical content selection because you only have to join or qualify using the single key (presumably perms etc were checked in the first query).
I looked at it for use with Greenpeace and decided to write a caching db_multirow instead.
I think we should look to a better solution than the existing paginator proc but that doesn't mean you shouldn't necessarily use it.
