Fortran especially. Forth seems to depend on processor characteristics. I seem to remember that Ada compilers routinely produce faster code than C compilers.
C has very loose semantics that make it difficult for a compiler to reason about code, as the saying goes, C is often little more than portable assembly. There's a reason that CLion was a big deal when JetBrains announced it, reasoning about C/C++ code is HARD.
If I wasn't more interested in Agent oriented languages, it might be interesting to see what could be done to design a new language built for performance. I would imagine you'd start with the Ada side of the house and work your way back to more pleasing syntax.
Ada and Rust are both significantly easier to perform aliasing analysis on, so frequently you can get much better code. Both languages have slight overhead for bounds checking and such, but you can turn it off (at least in Ada) if it's a problem (hint: it isn't).
Same reason Fortran is fast actually: it just disallows pointer aliasing entirely¹, meaning you get none of the flexibility of C pointers (heck, you don't even have pointers, basically), but if you're multiplying matrices it flies.
¹ I recall newer Fortrans have pointers, but as I'm not a Fortran programmer I don't actually know.
