Even in the '90s, Diablo was a mix of C and assembly for the bits that C was too slow for!

I've heard it said that, if they had the C compilers of today back then, the C code might've been good enough -- hardware being constant -- for them to have totally ditched the assembly, but I don't know how true that is.

Yes, but those C compilers of today run on the hardware of today! You wouldn't have been able to compile those C compilers to the point of efficiency required to build them in the first place. Lots of technology is like that. In order to do VLSI you need to do LSI, in order to do LSI you need MSI and so on all the way back to thin film circuits and then individual components. You can't normally skip such steps it's all incremental improvement on what's already there. In software it isn't much different and that software rode in lock-step with the hardware. The best I could do on a relatively fast micro of the 80's in terms of control, expression and eventually execution speed was a very nice macro assembler and given the speed of the storage devices and the raw clock speed of the CPU that was an accomplishment in itself. It pretty much maxed out the hardware. In spite of using clever stuff (hash tables for symbol lookups for instance, which ate up some of that precious memory, RAM disks etc) it still gook a good bit of time to assemble a 16K ROM image.

But then I got an Atari ST with a whopping 1M of RAM, a 10 MB hard drive and Mark Williams excellent C compiler and it was off to the races. That was a real game changer for me.

And many/most of them were subsets of the C you'd get on, say, a PDP-11, at the time (although I understand they got better over time). That was what always put me off using C on the 8-bitters. Turbo Pascal on the other hand.......