I'm an author of glide wrapper for other cards, but since some banshee users asked me for this. I've started to investigate about.

My idea is a Glide3x -> Glide3x wrapper with direct calls for funcs non affected by second tmu and a software layer for affected.

2 TMU memory loads: i'm thinking in split TMU memory in two parts and report only half memory at grGet

grTexSource: It Must be delayed for second TMU saving it in a variable.

Filtering: Are direct only need change texture address.

Polygon Draws: I want use multipass following next steps:
- save combiners status
- set first texture & effects (combiners)
- draw first poly (point, line, tri, quad)
- set second texture & effects (combiners)
- draw second poly (point, line, tri, quad)
- restore combiners status

SST/LFB/DEPTH: No changes

This is what i've concluded with all my investigations...
I'm going to start project but since i dont have a 3dfx card full compatibility with them is very difficult. For this i'm looking for a another developers.