sorry for that, I admit I did make assumptions in my responce, and I appologize.
I got vertex blending / weighting confused. I do remember reading that quake 3 took andvantage of hardware to reduce load on the CPU when it came to animation, I guess I assumed that was it using blending.
and I did toy with it a while ago, and my impression was that it was faster. (although my CPU is far from stellar - being a lowly 500mhz)

I still believe HL''s method of backface culling (which it does do in software it''self) would have been considerably faster on the hardware of the day (especially on the originaly release dates target hardware - ie, 2mb cards)...
There is also a vertex culling extension that works on the same principal, and although my machine doesn''t support that, from what I have heard, it does boost performance (and I''ve seen some modern D3D games show artifacts of this technique, so I''m guessing D3D supports it in some way)

HL''s frustum culling is a bit more extreme though, as the moment a triangle is off screen, it isn''t sent to the renderer.. which is probably a lot slower overall.

luckly, as I said, this isn''t true for HL''s model renderer.