It would certainly make sense. The last remaining thing for WM to do is to shift to GPU acceleration, which is ideal for heightmap manipulation.
As a side note, I’ve rolled my own terrain generator using OpenCL and its fast as hell. I’d love to help out doing any OpenCL porting 
Totally agree. Also this: http://runtime.bordeaux.inria.fr/StarPU/
I was also told that theres developments in gpu compiler stuff going on right now: LLVM with gpu backend.
monks
OpenCL would make a lot of sense. I’ve been playing around just slightly in this area, but I haven’t been able to put any significant time resources into it to determine feasibility. Everything does not need GPU acceleration, just a handful of the time-sucking devices – chiefly because we hit practical memory limitations pretty quickly.