> I wish the demo video was longer and gave a better idea of the gameplay, though. I might throw the $7 at it anyway to find out.
Just spent about an hour and a half playing with it. What isn't obvious from the video is that the resources at each compute node are very constrained. You only get one accumulator register, and one additional "backup" register that isn't directly addressable. Furthermore, the size of each node's program is limited to 15 instructions (and comments count toward that limit, for extra evilness).
The result is that the learning curve is pretty steep. For the first few levels, the problems are simple enough that you have a lot of flexibility in how to implement a solution. But once the problems get more complicated, you have to start getting creative in how you split the work across processors. It's actually more comparable to designing a pipelined CPU than it is to assembly language, IMO.
I just spent over half an hour coming up with a solution for level 7. It's a pattern-matching problem that superficially looks very simple: on a real computer, you could easily solve it with a finite state machine. But here, that program is considerably too big to fit into a single node -- and state machines aren't known for being easy to parallelize. I finally solved it by (rot13'd for spoilers) znxvat guerr pbcvrf bs gur vachg fgernz, srrqvat rnpu pbcl gb n abqr gung frnepurq sbe bar bs gur guerr cnggrea punenpgref, gura pbzovavat gur erfhygf jvgu gjb vafgnaprf bs na "naq" tngr gung unf n qrynl ba bar bs vgf vachgf. I ended up needing 297 cycles, which is about in the middle of the histogram; clearly there are more efficient approaches, but I haven't figured them out yet.