The same way as GMail and other large email providers. From my experience with GMail, its accuracy is near 100% and their heuristics is mostly based on statistical data, e.g. they probably detect large volumes of mail coming from one IP and having similar links, etc.

Yes. One of the nice things about a smart hosted gmail system is that when spam happens, they will see a corresponding flood of spam reports coming in over a certain amount of time, so that would make it much easier to stem the flood of spam. I am sure there is quite a lot to it which we will never know, but it does seem to be the best out there.