It's true that for extremely simple programs like the author's sole microbenchmark, non-C languages can get a speed boost over C from reasoning about aliasing.

It's true that for extremely simple programs like the author's sole microbenchmark, non-C languages can get a speed boost over C from reasoning about aliasing. But this doesn't scale. The author's argument breaks down as soon as you ask for an example benchmark that is larger and smells like something real.