polygons are simply composed of points in space with a surface drawn between them. With a series of xyz coordinates, a polygon can be drawn pretty easily.
NURBS and SubDs on the other hand use much more complicated math to draw on the screen.
Because games are real-time (that is, the action on the screen happens as you push a button and isn't rendered or anything) they need to be able to compute locations of objects, particles, run scripts, etc etc as fast as possible. Polygons are the most efficient way of doing that at this time. And I don't really forsee that changing any time soon. Even though machines are getting much more powerful, we simply increase the polygon count that they can generate per second, rather than trying to create things out of smoother geometry types. If we did suddenly try to make a game with nurbs, even with the next-gen hardware, we'd have to go back to, like, Nintendo 64 kind of graphics due to the computational math involved.
Hope that makes sense.