Before somebody else beats me to it, I have the % faster wrong. It should be written like 6.59x faster, not 659% faster.
I was a little surprised just how slow some things got, so the testing took longer than expected. I originally wanted to use 500k tracks, but iTunes beat the idea out of me.
They use a strange database, because it looks like there are two copies -- a native one and an XML one. Both have huge overhead for seemingly simple things. For example, making a few 100k song playlists cause the database on disk and the memory usage while running to bloat by over 100 MB.
They do a pretty good job at panes and search. We're faster, but not by orders of magnitude. However, I think we're doing a much deeper search so that it's not an apples-to-apples (or apples-to-jriver) comparison. We also have to support flexible grouping and expression based grouping, which isn't possible with iTunes.
One final thing that struck me strange was that to change a value I see in a list, I have to right-click and pick "Get Info" and then find the value in a popup dialog inside several tabs. Editing with a command called "Get Info" is wonky.
I may give WMP 12 a shake next.