Well, they're not, they're just an efficient way of encoding a subset of data. They can't compress that data, like, for example, compressing 100 x "a", unless they have a compression algorithm attached, they will just represent "a" 100 times, but more efficiently than unicode or whatever.