"We should forget about small efficiencies, say about 97% of the time: premature optimization is the root of all evil. Yet we should not pass up our opportunities in that critical 3%"
Unfortunately everybody seems to forget about the last bit.
One or two micro-optimizations may not add up to much. But a habit of, say, avoiding unnecessary object churn, might add up to a very large efficiency gain, a few microseconds at a time.
Admittedly this may depend on the domain you're working in. And of course some people may take the advice to an extreme in the other direction but this is (IMHO) rarer and much easier to correct.
Performance is a design issue and it is hard to do as an afterthought. In fact often microptimizations are the only optimizations you can do without changing your design.
That said I would draw a distinction between implementation and design. The surface API needs to be well thought out before hand. But this should be done before writing any real code (exploratory code aside).
(from comments of https://softwareengineering.stackexchange.com/questions/8008...)
value out > value in
That's similar to my reaction to "Unit testing won’t help you write good code." Of course it won't... but you are just using it as an excuse to not test you code at all.