yes, noise reduction using temporal / motion information is not really a new thing, many video denoising plugins use these techniques - but I'm not particularly surprised that no one really implemented it in a renderer. First of all, you can simply use a plugin in your postprocessing app, it does the same thing. And second, you have way more information at hand when you render an image than with just a series of single pictures from a camera - so there are better and way more useful methods to do it: you can preserve image detail by taking diffuse, reflection, lighting, etc passes into account. That's presumably also the way altus does it. Afaik you need a range of different passes, from normals to diffuse information to be able to preserve those features and using the 2 different qmc seeds you are able to identify the "real" noise and eliminate it without blurring texture or bump detail.
