The way I've used (and would recommend using) this idea is to spend some time deciding whether any big optimization is premature before implementing. By definit

The way I've used (and would recommend using) this idea is to spend some time deciding whether any big optimization is premature before implementing. By definition, premature optimizations cause more problems than they're worth, but appropriate optimizations pay dividends. You're absolutely right that the 97% number comes from a different time, but the central phrase is approximately as true as it ever was. The trick is that it can be really hard to determine whether a given optimization is premature. Like "one of the hardest things a working programmer will ever attempt" hard. One more thing: We optimize for factors other than CPU time, and all of the above still applies.