As I understand it, there’s a difference between “per clock” and “per pass”. The “clock” number is the number of textures you get “for free”. The “pass” number is the number of textures you can composite together before bunging them out to the framebuffer - better performance (less memory bandwidth required) and more blending flexibility.
Snippet from the NVIDIA performance FAQ:
“Although performing 2 texture multitexture may be more expensive than a single texture on GeForce 256 and Quadro, it is still much cheaper than performing another pass. And under most conditions, GeForce 2 GTS can perform dual-texturing at the same rate as single-texturing.”
I guess the same holds for GF2, so although 4x multitexture is slower than 2x multitexture, it’s still much faster than doing 2 2x passes and blending in the framebuffer.
Hope that made some sense… man, my writing sucks tonight…