There are already free voxel editors out there... not sure about polygon->voxel conversion tools.

Towards the end of the video, they say that they're not really using voxels. Polygon->voxel conversion shouldn't be any harder than scattered point -> polygon conversion. I'd say it's easier.

From some of their renders I'd assume that:

a) they found a way to quickly reconstruct geometry from scattered 3D point data (non uniform) - more info on a similar (but slower? approach) at http://hyperfun.org/OBS_AdaptCSRBF_SMI04.pdf

b) they have a sorting (search )  algorithm that pretty much allows them to easily load only small subsets of the of the 3D points depending on 'camera' view and position/object orientation/perspective/occlusion so that they only have to reconstruct/render the visible geometry

Shading won't work properly (and it doesn't in their videos) since hidden geometry can also contribute to the overall shading of a scene. As a workaround, they may be able to generate/use 'lower quality' models (smaller sample sets) for hidden geometry - but that will still slow it down a lot and invalidate their statement regarding the fact that they only need to process a number of points that's only up to the number of pixels on the screen.

While I don't see this technology make its way into games very soon (maybe just partially and for some VERY static objects like terrain), thank you svuun for bringing up this topic - I find it very interesting.

http://hyperfun.org/wiki/doku.php?id=hyperfun:main - a good place to start looking into 'non polygon' things