Quake perspective corrects only every 16 pixels and does linear interpolation in between. This is because the correction requires an expensive FPU divide.
While the FPU is busy doing that, the CPU can do the linear interpolation. Once the CPU is done with the 16 pixels, the FPU has also completed the divide and the CPU can do the next 16 pixels, and so on.
It was so long ago that I honestly don't remember. It might have been something like HL model viewer or Milkshape 3D rendering things using a different software renderer. I did some research and there is also an option to set affine texture mapping in the HL OpenGL renderer. HL works on Macs so I will test that stuff out soon
I think by the time HL shipped, the engine had been so hacked to bits that very little was left over from Q1. Heck, i think they even ported over various bits from Q2.