The first XScales have a broken cache, and therefore run quite a bit
slower than they should.

I gather that fixed XScale processors are just about now reaching the
streets.  You won't see a speed difference until/unless the
cache gets enabled the way it ought to have been able to in the first
place; not clear if WinCE will do this or not.

And to get good XScale performance, you'll need to rebuild for XScale,
as running StrongARM code will tickle slowness in calling sequences.

In short, both hardware and software have to be fixed to get
the XScale to work at speeds it should be able to....
                                 Sigh,
                                     - Jim

--
Jim Gettys
Cambridge Research Laboratory
HP Labs, Hewlett-Packard Company
Jim.Gettys at hp.com