Optimization in the Twenty-First Century

"But, uh-oh, strings are immutable, so each of these append operations causes the entire multi-megabyte string to be copied."

This is only true when the scope is limited to specific languages… I am pretty sure that the author is talking bare or low-level stuffs, but it would be better to mention some language scope…

I know people often repeat that -Os is often better than -O2 or -O3, and I try this regularly, but I don't think I've every seen a large program for which this was true. Does anyone have an example?

The very last point is the best, IMO. Look for ways to add concurrency rather than seeking faster sequential execution.

Easy to say, but hard to do right. Especially if you have shared state between execution contexts. Concurrency adds a whole new level of complexity and is hard to test and debug.