How about a 2-D test? What matters more to me is performance in "flat" applications. I used BitBlt when I needed to compare Matrox cards to integrated graphics at high resolutions.
BitBlt by michael herf (http://www.stereopsis.com/blttest/)
Savage IX on T21 (600 MHz) (XP)
BitBlt: avg: 113.6 fps [133.1 MB/sec]
ReverseBlt: avg: 5.1 fps [6.0 MB/sec]
Radeon on X22 (800 MHz) (XP)
BitBlt: avg: 186.9 fps [219.1 MB/sec]
ReverseBlt: avg: 5.0 fps [5.8 MB/sec]
Intel 855 on Motion M1400 (1100 MHz) (Vista "stark")
BitBlt: avg: 1129.1 fps [1323.1 MB/sec]
ReverseBlt: avg: 76.3 fps [89.4 MB/sec]
Intel 915 GMA 900 on X41 Tablet (1500 MHz) (XP)
BitBlt: avg: 1543.9 fps [1809.3 MB/sec]
ReverseBlt: avg: 69.1 fps [81.0 MB/sec]
Intel 945 GMA 950 on Motion C5 (1200 MHz) (Vista "stark") (2011-02-01)
BitBlt: avg: 1085.5 fps [1272.0 MB/sec]
ReverseBlt: avg: 1075.0 fps [1259.8 MB/sec]
Intel 945 GMA 950 on Motion C5 (1200 MHz) (Vista Aero) (2011-02-01)
BitBlt: avg: 909.8 fps [1066.1 MB/sec]
ReverseBlt: avg: 925.2 fps [1084.2 MB/sec]
All of these numbers are average of a few runs. BltTest is an old test, and finishes in a tiny fraction of a second on modern hardware, so its results are not steady. It surprises me that the Intel 855 and 915 integrated graphics both have higher "forward" speeds than the 945. Also surprising is that the 945's GMA 950 seems to have higher reverse speed than forward! I doubt I can trust this test with hardware made since 2004 or so.
(edited to add results)