This is a first cut, there are 50 simultaneous creatures and I want to render a few hundred, and I have a unique (if not really that practical, but maybe important for other tasks) way of doing it.
Processing source code is here:
It very slowly renders a single creature animation, while saving the frame and the entire depth buffer at the same time. On later passes it loads the old depth buffer and manually decides to use the new or old rendered pixel value- all the disk access and byte to float conversion code takes a long time.
I'm using odejava but it really isn't doing much except offering some joint and rigid body mechanics. Odejava seems extremely brittle, crashing all the time, the mechanics here are probably easy to duplicate since collision detection isn't needed.
Music is the cc licensed "Independence Day" by DoKashiteru featuring Colin Mutchler ccmixter.org/files/DoKashiteru/21426