Building from scratch without prior experience in the platform is commendable, and your write-up will help people experimenting along the same lines.
I watched the video and didn't realize how much of a penalty there is with painter's algorithm -- there is a very disturbing amount of overdraw, nearly 5x the amount of actual pixels in the frame. Did you give thought to raycasting the vertices of the quads to quickly eliminate polys that don't need to be drawn? A 486 has way more system memory speed than video ram speed, so eliminating overdraw is worth the effort.