Don’t attempt it unless it’s really going to be necessary.
Computers are fast. Really, really fast. They can do billions of operations per second. Shuffle gigabytes of memory in a flash. Stream high-resolution video without breaking a sweat. It’s quite impressive what computers can do today.
As a result, one should generally resist the urge to spend a lot of time on optimization and performance tuning. In most applications, no one will notice whether an operation takes three milliseconds or 300 milliseconds. So if you spend 100 man-hours improving speed and no one notices, then you’ve wasted valuable time that you could have spent fixing other bugs or adding new features.
Tuning for performance can also break things in unexpected ways and take a lot longer than expected to fix. Often a system that works perfectly fine will suddenly develop new bugs as programmers make changes as part of some unneeded performance tuning. So before you re-engineer a system to be massively scalable and ten times faster, make sure the gains will be worth the effort.