virtualbox.org

Actually I went back as far as 2.2.4. Installed a plain vanilla XP without IO/APIC (let call it "No") and one twin XP with IO/APIC (let's call it "Yes"). Nothing else enabled. The No installation was consistently faster compared to the Yes (~70%). I then enabled IO/APIC to the No guest and there was no change. At all. So it is NOT the IO/APIC in VBox. It is the difference in the Device Manager -> Computer, ACPI implementation of WinXP. Note that I did not increase the number of processors. Both single CPU.

I cannot verify on real hardware if changing the "driver" in the device manager causes that as well. But after an extensive multi-day research, I'm inclined to say that this may not be after all a VBox issue, but a WinXP issue. At best (worst) it could be the interaction between the ACPI setting in WinXP and VBox. I have no clues about that, nor do I want to speculate.

socratis wrote:At best (worst) it could be the interaction between the ACPI setting in WinXP and VBox. I have no clues about that, nor do I want to speculate.

If I was to speculate, I'd say that this should be the case, since having a multi-CPU winXP VM with ACPI setting doesn't drop performance in a Parallels Desktop setup, and winXP sees all 4 CPUs as expected.

So if you compare two 1-CPU WinXP VMs in Parallels, one where the computer property for ACPI has been modified in the Device Manager and one that has not, you do not see any differences?

Will try it and let you know. I suspect there will be no difference, since the same zip file takes around 1% longer to uncompress in Parallels VM when compairing VirtualBox VM with 1 CPU and Parallels VM with 4 CPUs. The Parallels VM has a few extra software (including antivirus software) which explains the slight delay I guess.