I think the author misses the point: you can't just look at code and expect to understand performance, in any language. That's what profilers are for.
I will agree that micro-optimization of <i>known</i> hotspots is mechanical, but finding the right areas can be deliciously difficult.