I'm led hear because java_grinder for the Parallax Propeller just got a mention on the Parallax forums. For the first time I think:
http://forums.parallax.com/discussion/168101/propeller-java. An impressive piece of work.
I don't know if you are aware but there is a way to have a very small, tight, PASM loop execute large PASM code from shared memory. Basically it fetches instructions from shared memory, one at a time, and executes them in COG. This is known in the Propeller world as LMM mode. Large Memory Model.
The C compilers for the Propeller generate LMM code so as to allow for reasonably sized programs. Despite the fetch, execute loop they do run at a usable speed. They can also, optionally, generate COG code of course.
GCC is extra clever optimization in that it has a thing called fcache. It will compile tight C loops into COG code and load that into into cog at run time. With fcache your Mandlebrot demo would probably run as fast in LMM mode as it does all from COG!
Anyway, just thought you like a heads up on the Parallax forum post and wondered if you would be interested in getting java_grinder to generate LMM code.