Famously, he also strictly enforced his simplicity also on the compilers, by insisting that e.g. things like optimizations should "pay for themselves" b

Famously, he also strictly enforced his simplicity also on the compilers, by insisting that e.g. things like optimizations should "pay for themselves" by speeding up the compilation of the compiler itself more than the extra code in the compiler would slow it down. This seems like a heuristic that breaks down as soon as the work a compiler does is sufficiently different from the work its output is expected to do. For instance, transforming calculations on vectors of doubles into SIMD instructions would likely never pass...