Aaron,

> >   As I wrote in an email, this can be quite possible because the
> > bytecode range check is not omitted in the interpreter loop for WinCE
> > VMs.  I don't know the snappiness you're feeling, but this is be
> > interesting.
> 
> I'm not sure where it is coming from either.  The benchmarks I cited in
> the list message I just sent out have the Zaurus 5500 at 22.5% faster in
> bytecodes/sec than the iPAQ or J720, but it feels a lot faster than that
> 20% margin.  When I tap on the background, it takes at least a full second
> to bring up a World menu in either 2.8 or 3.4 (even longer) on the WinCE
> machines; on the Z, less than a half a second.  It is quite surprising.

  One possible cause of this is that some "optional" primitives are
failing and the back up code is running.

  At one time, the stringHash primitive wasn't included in the WinCE
VM.  This turned out to be not a real performance hog, but anyway.  I
(or someone) should check out some other crucial primitives are
failing or not.

  However, I don't think this explains the poor sends/sec number on
XScale 400MHz.  Interesting.

-- Yoshiki