Yes, J. Carmack also mentioned that characters will have polygons for a while (especially with animations in mind). He also talked about some technical aspects I don't understand in every degree. But he wants to have some kind of sparse octree data combined with ray-tracing. Upcoming technology with multiple cores will be perfect for it (e.g. Larrabee).

The current C4 solution uses voxels for editing and as a data structure. But it also converts the renderable data into polygons in the end. So your graphic card still renders polygons. This way you have no problems with shadows. The standard shadow mapping applies at C4 terrain.

The advantage of such a system is: You can simply scuplt your terrain. You dont have to care about texture-mapping, you can paint textures on top of the geometry and you can have steep walls without distortions. You are simply faster as if you drag every polygon around. You dont have to care about polygons at all.
That is why ZBrush and 3d-Coat became so successful. They simulate the real artistic workflow (clay modeling) at our pc. They dont care much about polygons (at least not for the detailed object, we game developers still have to create low poly meshes).

And this is what the future will bring: more artistic freedom, faster workflow and thus more details in less time.