A lot of voxel systems are square for convenience sake, there's a few variations, I think those are point clouds, so more like little atoms, not fixed to a 3D grid like traditional voxels.
The rendering of the point clouds, well the USP of this system really is the way that it decides what colour the pixel should be. For the 'search algorithm' to be fast, it would rely on some sort of indexing, and the minute you do that, you limit what the system can actually do. It's probably great at rendering this data in a fixed environment, no movement, no animation blah blah blah. Something moves, and it affects the index, the index is probably massive as it is - the whole house of cards comes crumbling down. That's why this engine has very limited use IMO, there are probably game ideas that would work nicely, but it's not exactly the sort of games that people play in great numbers.
Either way, as people keep saying, movement is a major hurdle that they aren't properly tackling yet - and a video from another developer that is too slow to be usable is not a valid rebuff.
Personally I hold more hope for polygon smoothing technology, like using polygons but polygons that can be infinately smooth - like a bezier curve. That's the logical progression IMO, because it wouldn't cost the same data footprint - in fact polygon counts could drop substantially. Imagine a face model that is incredibly detailed and smooth, using a normal map to define the curvature of the polygons. Look what happend to voxel terrains a decade ago - we had the Delta Force games, using massive and detailed voxel terrains, running nice and smooth - now they are never used, and it's largely because they aren't necessary IMO.
Health, Ammo, and bacon and eggs!
