Supports Fast3D, F3DEX, F3DEX2
You can choose whatever vertex cache size, but I've only tested it with 32 vertices/F3DEX2
Run without parameters to see options.
If you make it better let me know.
BTW you can try messing with the per-vertex ambient occlusion. Just for shits I tried it on a model I ripped from the Fable silverlight demo, just hacked it into my racing engine
Also, for the love of god don't develop N64 using an emulator. That 15k model crashes the RSP, lol.