I'm running parallels 3 on a MacBook Pro (running OS X 10.5 Leopard beta) and find that hardware accelerated OpenGL performance is terrible. An application that runs at >30fps on bootcamp runs at <3fps under parallels.
I ran Cinebench, an OpenGL benchmark test. I tried it as a native OS X 10.4.9 app and as a Windows app under XP in Parallels 3.0. This data was originally posted in this thread, and only uses production software.
My Parallels numbers (OS Window mode only):
OpenGL-SW-L: 834
OpenGL-HW-L: 446
My OS X numbers:
OpenGL-SW-L: 2007
OpenGL-HW-L: 4211
The hardware numbers reflect almost a 10:1 degradation in Parallels, which might correlate with your 30fps:3fps numbers.
The really strange thing is that the Parallels hardware numbers are LOWER than the software ones.