I've been sidetracked for a couple of months trying to learn global illumination and sss techniques. I hope to merge it with the 3D volumetric fractals over the next few months.
basic GI and even SSS aren't too difficult to understand and implement (such is the beauty of monte carlo methods), however doing it
efficiently is where you'll spend years of studies.
for low-resolution volumetric datasets (anything on a grid will be low resolution), i think the best way to go would be photon mapping. i can recommend henrik jensen's book on the subject, it clearly explains the algorithm and how it's best used, and even provides sample code; moreover, the details of efficient volume scattering (something quite difficult to find in the literature -- i own many rendering books and this is one of the few that cover it) are presented therein.
http://graphics.ucsd.edu/~henrik/papers/book/monte carlo methods are somewhat ill-suited (complete overkill) to a problem like this, where you have an explicitly bandlimited volume density given by the grid resolution.
