The first is the number of processes that the SkunkWeb server spawns. If you are running into a performance plateau and your CPU utliization is less than 100%, try increasing the number of SkunkWeb servers (by changing the numProcs config variable).
Another is the memory compile cache (activated by the useCompileMemoryCache option). What this does is keep memory images of the on-disk compile cache so that it can reduce I/O and CPU utilization that would normally be used loading and deserializing the compiled forms of compiled things (template, python code, etc.).
The big tuning parameter is by far: the cache.