I'm honestly a little surprised that ACPI has a Sleep()-like function at all, I can't really imagine many situations where the firmware actually wants the user operating system to wait for any real length of time, as that would block any other ACPI events, even if it wasn't in an interrupt handler. I feel there's pretty easy ways to deadlock the system in that sort of situation.

And also surprised that windows actually allows the ACPI driver to sleep in an interrupt handler - on Linux that'll immediately BUG()... Unless windows doesn't and the above ACPI blocking is what they're measuring here.