The often repeated mantra that high level language compilers do a better job than humans isn't true, and doesn't become true through repetition. The compilers can do no better than the person programming them, and for a finite size compiler, the optimizations are generic, not specific. And a good low level programmer can take knowledge into effect that the compiler doesn't have.
Two things. Programmers are expensive, silicon is cheap. And really good low-level programmers (which I am not and don't claim to be) are (and deserve to be) very expensive indeed.
When you're writing a bit of code which is going to run on hundreds of thousands of processors all over the world (as OpenSSL is), the cost/benefit does move in low-level programmings favour. If you save just a few cycles on each of billions of operations on millions of processors, it's worth throwing resource at that optimisation. But OpenSSL (like the Linux kernel) is a special case. Most perfectly ordinary good programmers will write more cost-effective code using languages which don't require them (for example) to keep track of memory assignment.
Also, most human beings can't hold an effective map in their minds of the operating states of a moderately complex program. And people make mistakes - even the best of us.Software, by contrast, is very good at detail, repetitious, complex tasks like keeping track of what bit of memory has been assigned for what purpose, and noticing when nothing remaining in the system holds a pointer to that data. There are whole classes of programming error which good compilers will simply never make.
I was doing a post-accident audit on safety-critical (closed source) code a few years ago. The reason for the failure was that someone had used strcat to concatenate strings for an error message, where the strings being concatenated were stored in the data segment (yes, think about it for a moment). It cost millions of pounds worth of damage, and it was very lucky no-one died. If the software concerned had been written in (for example) Java, the accident wouldn't have happened. Yet the software had been written by a very senior C++ programmer, and had gone through four separate code reviews before being accepted into service - and all four had missed it.