REXML performance gap between XP and OS X

I usually don’t notice these things as the differences are usually
irrelevant, but this one has triggered my curiosity:
I get a significant diference in performance using REXML betweeen OS X
and Windows XP. And with significant, I mean a factor of 4 with Windows
beeing the slower.
I expected XP to be slower, since I’m running a Parallels VM, but not
that much.
A bit of context: I have a 3.2 MB XML file and I am running several
XPath expressions over it collecting elements.

Two of these expressions are //element_name so that takes a bit of time.

Here are the results of a unit test on OS X
[20070906 16:00:39] INFO: Product groups parsed in 4.142405 seconds
[20070906 16:00:43] INFO: Activity groups parsed in 4.319004 seconds

and the same unit test on Windows XP
[20070906 16:01:49] INFO: Product groups parsed in 15.142 seconds
[20070906 16:02:07] INFO: Activity groups parsed in 18.497 seconds

Windows runs of my BootCamp partition in a Parallels VM with Coherence
enabled and 850 MB of memory.
Even considering the drop in performance the VM incurs I still think a
factor of 4 is a bit much.
Results have been consistent, meaning repeated runs give results in the
same ballpark (depending on what I am doing at the time the unit test
runs things run a bit faster sometimes).
The windows-based unit tests exhibit a worsening of performance as the
load on the CPU increases which is not linear to the worsening of the
test’s performance on OS X:
on a clean system with just the VM loaded, the difference is a factor of
~3.5: ~2 seconds for the OS X test per parsing and ~7 seconds on XP but
I can attribute that to the tuning of the VM which favors OS X

Any thoughts? Can it be that the VM influences performance so much?
If I get the time I’ll boot BootCamp and see what native XP does.