Not really, the author's are bench marking on different parameters to yours. His is size and yours is speed.

Except if the algorithm is changed. On https://codegolf.stackexchange.com there are regularly solutions that are very short, but by necessity also employ a suboptimal algorithm, and are thus slower (sometimes very much slower) than the optimal program in terms of performance.

How many instructions is an infinite loop :p More seriously, processors are superscalar and can execute out of order.

In Intel, basically one.

Not at all true. Especially with SIMD-type instructions. Lots of verbosity and boilerplate. But the perf is undeniable.

Allen's original paper includes loop unrolling as a useful optimization even in the 70s, and there can be good reason for inlining.