Adding my $0.02 to the well-done approaches to the problem above:

The data looks so much like XML that I am tempted to convert it to XML, then use one of the many XML parser modules to extract from it.

I'm too shameless, and lazy to actually develop and post code, particularly on a Sunday.

I feel that the XML approach is scalable, and easily extendable to help parse the majority of Win32 performance info.