on compilers that support it (notably: gcc, SunPro, icc), the bytecode
evaluation loop is compiled with a new dispatch mechanism which gives
speedups of up to 20%, depending on the system, on various benchmarks.