Many 3D cards have higher fillrate with disabled depth buffering.
The difference may be as low as several percents or may be very significant.
For example, for 3Dlabs Permedia2:
83Mpix/sec (without depth buffering)
43Mpix/sec (depth buffered)
And some 3D cards prefer front-to-back order when depth buffering is enabled.
I remember it about Rendition V1000, and now looks like ATI Radeon really like it.
But polygon sorting may lead to frequent texture/whatever changes.
You can create sorted by texture “renderlist” when you walk BSP tree.
And you can create additional sorted by depth list with all transparent polygons.
The renderlist may contain opaque polygons, sorted by texture and F->B for each texture, and transparent polygons, sorted B->F.
[This message has been edited by Serge K (edited 09-15-2000).]