[koala_man]

This kind of CS-based rationalization is arguably another aspect of what the article comments on. I wrote a benchmark and found the difference in this case to be 3x-3.5x.