We've seen people do some pretty awesome stuff with the Kinect that expands the peripheral's applications beyond the realm of controlling games, but we've yet to see anything that matches the wide-reaching implications of Microsoft Research's KinectFusion HQ concept.
Using the Kinect's dual cameras, KinectFusion HQ can construct an accurate, high-quality 3D representation of an entire room in a matter of seconds. The longer the Kinect has to look at a specific object, the more detailed that object then becomes; able to accurately scan and render millimeter differences in texture.
Once the room is rendered, the model can then be interacted with in real time, which the above (oddly silent) tech demo demonstrates in increasingly wild and mind-blowing ways. Since the Kinect is typically thought of as a stationary object, it will be interesting to see how this technology may eventually be implemented into gaming, since its primary functionality relies on moving the unit around the space you're scanning.