Passing a function into a sort() is going to be slow in any language, unless the language is able to inline the function. That depends on static analysis by the compiler, which is notoriously difficult in Perl.
The key, as you mentioned, is to use builtins, or more generally to use canonical expressions. There's always more than one way to write an expression, but Perl definitely has opinions about the correct way(s), and the Perl compiler does pattern-matching against the code to recognize many of the canonical expressions. They then get compiled to higher-performing IL code. This is why using things like the spaceship operator, <=>, will run much faster than equivalent verbose comparison code.
This is over-simplified, but I always thought about it this way: the Perl compiler generates "assembly code" for the "Perl CPU" to run. It can generate verbose assembly code using lots of low-level assembly commands, but it can also recognize commonly-used expressions and compile them to a single high-level assembly command, which can then run in a single "CPU cycle" instead of stepping through verbose code. Therefore, it was always worthwhile to learn how to write the commonly-used expressions in the common way, so that the compiler can recognize and optimize them.
I don't know if other languages have this kind of optimization. JIT compiling, as I understand it, does a quick and rough compile without optimizing on the first pass in order to start execution quickly, and then recompiles with more optimization on code blocks that are consuming a lot of runtime. I don't think any of the JIT compilers look for common expressions and apply pre-determined, hand-coded optimized function calls to them.
The key, as you mentioned, is to use builtins, or more generally to use canonical expressions. There's always more than one way to write an expression, but Perl definitely has opinions about the correct way(s), and the Perl compiler does pattern-matching against the code to recognize many of the canonical expressions. They then get compiled to higher-performing IL code. This is why using things like the spaceship operator, <=>, will run much faster than equivalent verbose comparison code.
This is over-simplified, but I always thought about it this way: the Perl compiler generates "assembly code" for the "Perl CPU" to run. It can generate verbose assembly code using lots of low-level assembly commands, but it can also recognize commonly-used expressions and compile them to a single high-level assembly command, which can then run in a single "CPU cycle" instead of stepping through verbose code. Therefore, it was always worthwhile to learn how to write the commonly-used expressions in the common way, so that the compiler can recognize and optimize them.
I don't know if other languages have this kind of optimization. JIT compiling, as I understand it, does a quick and rough compile without optimizing on the first pass in order to start execution quickly, and then recompiles with more optimization on code blocks that are consuming a lot of runtime. I don't think any of the JIT compilers look for common expressions and apply pre-determined, hand-coded optimized function calls to them.