Nim, with the new ARC garbage collector works well for those situations now (well within a few percent usually). Especially with the `move` and `sink` semantics.

I'm using it in embedded devices and needed a few tight loops a week or two back. With a bit of work and a few odd contortions, I got it to compile to almost the exact type of C code I'd write by hand to run a fast inner loop using preallocated arrays, etc. It turns out that code wasn't noticeably faster than more idiomatic Nim code with a few tweaks to ensure move's were able to be used. So I ended up going back to the more readable/idiomatic version while still roughly doubling the overall speed.

The new `move` semantics with Nim's ARC system really do help make it easier to write fast code. Now I understand why the C++ ecosystem is so bullish on `move` in the C++ stdlib now. It really can take code which looks like straightforward C code but performs closer to painstakingly optimized C code. Mainly since it result in less copying of objects/structs and reducing the number of malloc's. Overall it means less effort to design a fast system.