Its reliability (not scaleability, as the question asks for, sorry) is pretty good but through our testing we know that some clicks do not get recorded. It doesn't bother us a lot because we look at the aggregate of thousands of visits, but we do know it misses things.
As for scaleability, it's about right. You really don't want zillions of clicks per snapshot - the screen just turns to 100% dots and you lose the ability to differentiate different screen areas. We find that 25,000 clicks for a page gives us a really good view.