For games, there are two main reflection techniques.
You mentionned cube maps that is one of them. Cubemaps are good for reflections on arbitrary complex objects (sphere, teapots, ...). One other advantage is that they can also easely handle reflections on normal mapped objects.
In terms of computing poxer needs, you have to be aware that for each reflecting objects, you have to render 6 times you scenes to the cube map. There are several optimisation possible (not render cube map at each frame, use low def render surface, ...) but it remains costly.
The other way is to use "portal" related techniques that are very well described in this questionin this question.
The main advantage of this technique is that you will only require one more rendering pass. Another good point is that you can handle mirrors reflecting mirrors in a more complete and easy way than with cube maps (still it is not esay to handle such cases).
The main drawback is that you have to cull objects that are behind the reflective surface. Another drawback is that reflective surface must be a plane.
Raytracing in real time games is, afaik not used because it is too much resource hungry.
