With the help of Superman providing the link to the excellent stereo image viewer and converter here -
http://stereo.jpn.org/eng/stphmkr/ and Stefem for explaining how to take stereo screenshots using the nVidia 3D capabilities, I've been able to compare nVidia's semi-automated 3D results to those you can get using the manual screencap & edit approach.
My results aren't definitive, and with some tweaking of the nVidia 3D parameters, maybe these results could be different, but...
Cliffs Of Dover and nVidia 3D are clearly having some issues.
Manual
nVidia
Manual
nVidia
Manual
nVidia
Manual
nVidia
Manual
nVidia
Manual
nVidia
Manual
nVidia
Manual
nVidia
Manual
nVidia
Manual
nVidia
Manual
nVidia
These nVidia screenshots were taken in PNS format, using what looked at the time to be approximately the same amount of camera offset as the PNG FRAPS captures, but the 3D effectiveness is quite different. This might be a result of adjusting some other nVidia settings incorrectly, or it might not be, I don't know. I'm assuming there must be some way of adjusting the nVidia 3D parameters so that the basic 3D effect is more pronounced, because if there isn't then what's the point?
I don't think the bugged out water and shadows on the nVidia shots is the result of an end-user setting though.
I was hoping to see if I could use the nVidia 3D, then run the game in slow motion and take repeated 3D screenshots that I could then compile in a video-editor as a moving image sequence. That would have meant not having to use the game's buggy replay system and manually resetting the camera position for every shot. But I guess there ain't no easy way out
Maybe some of the other nVidia 3D users can confirm these graphics bugs I'm seeing and perhaps suggest some solutions.