I just found out that the gnuify script was not up-to-date, the formatting
of the generated interp.c had slightly changed. This caused the bytecode
interpreter loop not to use registers. Same for the primitive dispatch.

On my machine I get a 25% speedup from this:

0 tinyBenchmark went up from 
	'45325779 bytecodes/sec; 1455887 sends/sec'
to
	'55555555 bytecodes/sec; 1636262 sends/sec'
Get it from CVS at squeak.sourceforge.net :-)

Now, is this still slower than the Windows VM? This is a PIII/600.

-- Bert