Stefan,

> Well, the "1 tinyBenchmarks" numbers are quite impressive.  I 
> get about 12% speedup in message sends and 70% in bytecodes.
> 
> However, when I do a "Time millisecondsToRun: [1 tinyBenchmarks]", 
> the old VM is about 0.2 seconds (or 4%) faster, that is 4.8 instead 
> of 5 seconds.

The absolute time spent in #tinyBenchmarks is absolutely meaningless.
#tinyBenchmarks calls #benchmark and #benchFib repeatedly to achieve a
halfways accurate measurement so the difference you're seeing comes from the
different number of iterations.

  - Andreas