Static mesh performances

[skarab24]

I'm planning for the next "awesome" feature of Coffee.
I would like to know if adding static meshes rigid bodies won't kill to much, in that case : I'm planning to add full marching cubes support, this will be used in real-time, for terrain and indoor levels creation as well. CSG is good but I haven't the manpower to support both. I will have full LOD support as well (I already have it in the core, but it's hidden and not tested for now), but maybe the lowest LOD (or nearest), will be used for physics, depending of perfs, and colliding issues (read: switch physics mesh between two Physics simulations). I'll have automatic mesh LOD generation as well, using QSlim for user meshes (read: edges reductions with possible gaps).
To note, I'm using google code synchronized as often as possible.

Is there some pitfall I should be aware of ?

Note than I don't target Leadwerk/Unity/C4 Engine performances, which I plainly respect, but I need the whole thing to be real-time, and it's going toward Open Worlds...

[JoeJ]

I plan to do something similar in the far future in case i'll store terrain in a volume.

Julio, you talked about voxels and signed distance fields recently.
Would it make sense to generate that data ourselfes und use it for that purpose,
meaninig it handles also rigid body collision, not just cloth and soft bodies?
I'd like to see it as an alternative to heightfield, as it should be faster than static mesh.