Fusion-icon starts compiz with a couple of options, which might have that effect...
We're now talking about 2 different problems here: slow compiz, and slow 2d.
The gf8400m GS is just slow. About the same speed as a gf6200, or a gf Ti4x00. If you thought that was a high-end chip, you were severely misinformed, but it should be fast enough for simple 3d and all 2d...
Anyways, first problem: it should be fast enough for compiz, except for the heavier effects(transparent cube, blurring, fire, etc. ) I've had a gf6200 doing just that. A couple of options might help though: disable vsync(I'd rather have a bit of tearing than a stuttering desktop), don't set the framerate too high, manually set the screen-refresh-rate in ccsm(nvidia's DynamicTwinview-function makes it report 50Hz) and use conservative options for all the plugins. Furthermore, you can configure the polling-frequency for powermizer in nvidia-settings(last page). Set that to 12ms or so.
I myself have been using Kwin4-compositing since kde4.0, which is stable, fast, and just got wobbly windows. Before that, I've used compiz for about a year, and could never really like it. It is still pretty meagre as a window-manager (I did upgrade my pc at the same time, so I can't really compare performance.)
second problem(slow 2d): even if nvidia doesn't say anything, doesn't mean they're not listening. I've never seen them really respond to 2d-performance-problems, but new drivers do tend to fix at least part of the problems(first 100-driver got fully-accelerated subpixel-rendering, they got xrandr-acceleration, 171 got glyph-cache,...). I myself am using kde4 on a gf8600, and as soon as it starts doing some real 2d-rendering-work(smooth-scrolling ,heavy scrolling, xrandr-ed animations, and basically anything non-static), it does eat my cpu/becomes slow, so yes, I do agree there's a problem there, but on the whole, kde4 is quite responsive, and getting better, so it's something I can live with, especially considering the alternatives. I guess this is a hard problem to tackle.
My gtk-perf-scores(-ac 1000), on gf8600gt with nvidia-drivers-171.06, InitialPixmapPlacement=2 and GlyphCache=1, on gentoo-64, with an athlon-64-4400+-proc:
Kwin-compositing on, gtk-qt with keramik theme:
GtkPerf 0.40 - Starting testing: Tue Apr 22 04:36:45 2008
GtkEntry - time: 0.80
GtkComboBox - time: 45.35
GtkComboBoxEntry - time: 41.10
GtkSpinButton - time: 7.17
GtkProgressBar - time: 6.01
GtkToggleButton - time: 10.76
GtkCheckButton - time: 10.94
GtkRadioButton - time: 14.41
GtkTextView - Add text - time: 47.81
GtkTextView - Scroll - time: 15.10
GtkDrawingArea - Lines - time: 5.49
GtkDrawingArea - Circles - time: 6.30
GtkDrawingArea - Text - time: 4.86
GtkDrawingArea - Pixbufs - time: 8.95
Total time: 225.05
GtkPerf 0.40 - Starting testing: Tue Apr 22 04:41:54 2008
GtkEntry - time: 0.79
GtkComboBox - QObject: Do not delete object, 'unnamed', during its event handler!
GtkComboBoxEntry - time: 18.37
GtkSpinButton - time: 2.43
GtkProgressBar - time: 1.79
GtkToggleButton - time: 5.23
GtkCheckButton - time: 5.54
GtkRadioButton - time: 6.86
GtkTextView - Add text - time: 48.38
GtkTextView - Scroll - time: 8.36
GtkDrawingArea - Lines - time: 2.48
GtkDrawingArea - Circles - time: 3.19
GtkDrawingArea - Text - time: 2.82
GtkDrawingArea - Pixbufs - time: 5.20
Total time: 131.67
So, kwin-compositing does halve my rendering-speed, which I also notice in opengl-apps, so I mostly run those on a different X-screen. I also had that with compiz on my old gf6200(compositing does speed up moving and switching windows though, and, if well implemented, can result in a general performance-increase, but not with nvidia...).
BTW: Phoronix is busy building an extensive linux-testing-suite/framework, which can be found on http://www.phoronix-test-suite.com/index.php