The tip about “prefer pre-shuffled lookup tables to actually calling random[]” is a good one. For the “Screen Melt Transitions” deck (thanks for the shout-out!) I tried to generate random numbers at the beginning of each transition, and as a result I had to lower the transition quality to fit within Decker’s calculation limits.
I do like the way that the transition looks a little different each time you see it, but maybe I should try randomly generating numbers at module-initialisation time rather than on the first frame of the transition. I could even add a reseed[] method to call outside the transition effect, to generate a new pattern for the next transition.