The position in the matrix you mention is what move the vertices when you multiply them by the matrix. In the end, you have to move everything into a unit size cube (z-range differ between OpenGL and Direct3D though) centered at the origin. This cube cannot change so you must move the vertices into the cube, not move the cube to around the vertices. This is why documentation says you move the world and not the viewpoint. The two are conceptually identical though so it is a matter of how you look at it.
Now that makes a lot of sense. I always assumed they were different techniques.
Thanks very much.