That's a cool idea. I've been thinking of something related for a while.
I haven't acted on it because it's just so darn complicated, but you should be able to scan a 3D object with your G1 and a laser pointer.
If you combine the camera with an angle-mounted line laser pointer, you can get a single 3D data slice. If you move the camera and laser, you can get another slice... and so on.
If you know the path the camera traveled while it was recording slices, then you can piece together all the slices into a complete 3D model.
It happens that this is a somewhat solved problem. You can track a camera's path by identifying "corners" in the images and correlating them between frames to measure the camera's movement during that frame. The problem is that it is computationally very expensive - probably too much for your poor little G1's mobile processor to handle.
But wait! The calculations can be greatly simplified if you have a good guess as to which direction the camera moved... And the G1 has...
wait for it...
So the accelerometers can tell roughly where the camera went during recording and the image-based method can refine the motion from there - which is computationally feasible at that point.
Then combine the 3D slices into a mesh of triangles and... voila! You have a full 3D model of your hand, your computer mouse, your shoe... whatever. You can display it with openGL, send it to a friend, or save it for later - who knows.
Too bad it's so complicated though
. Probably too much to whip together in a weekend.