Think PBs for a game the size of Crysis. We have had 3D textures for a long time now, there is a reason we don't use them. If you are repeating one identical model or a small set of them, not a big concern- for wide open varrying areas? RAM far beyond what we have.
The models they show, as they show them, can't be animated- that's why they were all static(you would need to create a skeletal system then tie points to bones for every possible animation- exponential increase in computing resources needed)
Shader technology drops back to DX6- you can't have surface based computations if you don't have a surface.
Creating assets- Making a model of a tree would be trivial, the complexity of making moving objects in a real time environment would be a nightmare. This type of technology, although far more advanced, has been used for offline rendered images for a long time- but that is when the artists are moving things themselves frame by frame.
In terms of seeing something comparable using existing technology, scroll to the bottom of the screen and watch the video-
http://www.3d-coat.com/voxel-sculpting/
He can keep going until he runs out of memory. Nothing I saw in the Unlimited demo looks as good as what they are doing in 3D Coat, let alone better, and that is something that will run on your old 8800GTs you may have collecting dust in your closet.
Their search technique is simply a type of ray casting(whatever they may want to call it) that PowerVR has used in commercial products for years(move along, nothing approaching new to see here).
On a realistic basis if nV or ATi wanted to they could be pushing billions of polys per second without a problem, that isn't where real time 3D needs to be pushed the most at the moment. The big advances are coming in shader technology and that is an area this new engine would suffer badly at, along with animation and memory amounts.
I'm not saying what they are claiming wouldn't work, I'm saying it would not work nearly as well as what we already have.