Not to limit the number of polygons on screen, but to make it feasible to precompute the exact order polygons would appear on the screen during each frame at render time. Pre-computation actually increased the effective polygon count because it also meant that the engine didn't need to render occluded polygons (i.e., the rock behind the tree you're looking at) at all.

is that similar to resident evil's prerendered graphics?

Resident Evil largely baked their complex scene geometry down to large, flat textures on large, flat polygons in game. Crash kept the complex geometry of the scene in the game. What was baked was the exact set of polygons that needed to be loaded and displayed at any point along the track.