Comment by Spikefu

Not necessarily. I just finished optimizing an AES implementation where some of the functions are called hundreds of thousands of times when decrypting a 300k payload. In cases like that, micro-optimization is not at all pointless unless you forget to benchmark the whole thing for each version and compare.