TreeItem expand/collapse performance in hardware pipeline is worse than in j2d pipeline.

Here are results of running "TreeViewExpandPerfTest -size 100 -type wide" with fx2.0-b39 on windows xp midrange machine.
 hw: 34.92 fps
 j2d: 51.56 fps

I also see similar on my Dell E6410 laptop.

It would be good to investigate and improve.