This doesn't actually fork a JVM process, right? It's just splitting up a calculation "unit", which behind the scenes gets mapped to different execution threads. Does that sound about right?
Argh, can we please stop using exponentially recursive Fibonacci as an example to parallelize? Because using a linear (or even constant-time) algorithm makes it moot.
http://hawtdispatch.fusesource.org/
Another tangent: it's too bad java's servlet API is defined in a blocking style. Is there a JSR to remedy that deficiency?