There was a thread where someone was converting a Dinosaur (iirc) to triax weight mapping. After a bit of experimentation I believe they found that the best method for them was to be conservative on the weight mapping, then take the joint in question through a series of bends, smoothing out the gradient in areas where the bend was not smooth enough. Again, pulling from memory but I believe they found this much easier then putting in more to start and trying to reduce. The trick seems to be the direction of the workflow, conservative with adding.