Modern CPUs often keep a hundred instructions in flight, and accessing main memory once can take more time than a hundred simple instructions. Accessing main memory involves cache coherency protocol logic with neighbouring cores, and it may involve locking if the program wants a read barrier, a write barrier or any kind of volatile variable.

Instructions like CMOVBL involve only a small number of CPU registers, nothing else, and can't interact with instructions far ahead or behind them in the instruction stream, or with other cores/threads at all. Very little state. They're simple to reason about, both for the compiler authors, the CPU and the poor developer who's chasing a threading bug.