I started working on games in the early 90s and was one of the first to use C instead of asm because my game was targeting Amiga and PC and didn’t involve significant cpu intensive graphics etc. However the rest of the company was working on Super Nintendo, Megadrive etc and these were exclusively asm. C compilers back then actually did add significant overhead to code and you could do much better by hand, especially since cpus in consoles often lacked features such as floating point units or had cut down memory management units. In addition they had custom hardware that you could use (blitter for example on Amiga, gte on PlayStation) that was easy enough to use from asm and the compiler couldn’t help you.

The main advantage that Chris used in RCT2 to optimise his code was that he could vary his calling convention however he felt. Most of the time all variables were passed in registers instead of the stack. This can make for some impressive speed ups.