I see what you mean. I encountered the problem on SCOP and I usually store the result that doesn't need to be re-computed in a data blob during the first evaluation of the operator.
I'm done with my first test with ICE Kinematics. Basically what I'm trying to do is to create a percentage constraint on a nurbs Surface. For that I'm sampling the nurbs, let's say a hundred time, to calculate an approximation of its length. This sampling of the nurbs curve only needs to be computed once. Right now it's re-evaluated for each output... Well, I still need to work on it a bit. I'll post something here as soon as I can.. need to finish my production work first ;-)
