I think the engine could/should support variable or even infinite voxel resolution, through support for node subdivision.
http://www.minecraftforum.net/topic/128 ... xel-world/
I wrote about my vision of the gameplay side of this, a few years (holy shit!) back, and I don't think I can explain it much better than I did there. Of course it could vary greatly, go off in a number of different directions, which would be the great thing about the minetest engine/game split.
On the tech side, the simplest I could try to explain it, or implement it, would be to make each node potentially a reference to another block (whose nodes may also be blocks!). Of course this could easily get out of hand, but the trick to making it usable at all would be to also enable the engine to begin ignoring subdivisions below a chosen level (chosen by the game), and to make interacting with the subdivision levels dependent on special conditions (again varying by game). Subdivision levels that were being ignored, would be treated just as a node entire, and rendered as such (just like they are today - graphics generation or selection method would be game specific, of course). The game could choose to swap around and through the ignored subdivision levels at will, of course, for example to run active/simulation elements in them (at whatever level of complexity it decides, which could also vary - all this stuff would be game decisions).
I'm not sure if any of this is conveying, or if it just sounds like crazy talk. But I think it could really work, and would be the next step forward for the voxel 'genre' and technology. Both for very low-level/'high-res', but still technically feasible voxel games (not just building games? as a physics engine?), and for more traditional minecraft-like genre of block building games, where it could be used as I talked about in the linked post above.