Since we have such a large amount of feature requests our attitude is typically to implement the minimum necessary and move on to the many other crucial things we need to do, like better supporting mobile. Obviously we could implement circular collisions, but I think there's lots of other stuff that takes priority. I'll add it to the todo list though. When we first added collisions in September there was no physics, no Chrome Web Store, Facebook, PhoneGap or Scirra Arcade output, no Families, and so on... hopefully it's clear to see why we didn't spend ages on one feature when we had so much else to do, and that's still true today.
Using an eight-point octagon should be a good approximation though - or is there something about that which makes it unsatisfactory?