We present a new class of methods for variance reduction of Monte-Carlo simulations of radiation transport in inhomogeneous media and also present specific implementations from this new class. The intended application is cancer therapy dose planning although it is likely to find application in other domains. The technique takes advantage of the continuity equations for flux which underlie the transport. Instead of smoothing dose after a calculation, we smooth something which is proportional to the local scalar fluence by pre-scaling the data before smoothing, and then re-scaling afterwards. This allows true sharp edges in the dose, which result from discontinuities in the tissue (bone to soft tissue, for example), while allowing very aggressive smoothing of the fluence, which is a very smooth function. This allows multiple order-of-magnitude reductions in the computational effort to achieve a given level of statistical smoothness in a therapy plan thereby dramatically reducing the computational time requirements for full Monte-Carlo based therapy planning, making such planning routinely possible even with quite modest computational resources.