Since this thread has been revived...
I suspect Parallels pins down the entire RAM space of a running VM. If they didn't it would be eligible to be paged out to disk. The guest OS would have no knowledge of this: it would appear as though some areas of "physical" RAM were sometimes magically slow by many orders of magnitude.
When Intel launched HyperThreading Microsoft was forced to admit that Windows had many software timing loops so sensitive they would break under HyperThreading and that was minor compared to this!
It's possible that Microsoft only did enough of a bandaid to survive HyperThreading and that software timing loops that persist to this day would break using non-wired memory. And, some people will want to run older versions of Windows (perhaps old systems whose hardware has been physically retired but which may still need to be accessed from time to time and have been migrated to a VM). And there may be other operating systems with timing loops they'd like to support.
I do see an apparent leak bug in 3170 but in active not wired memory. Post on that coming shortly...
Last edited: Feb 18, 2007