The author doesn't seem to understand the birthday paradox. Any hash which gets fewer than ~12 collisions (for his dictionary and hash size) is doing so by pure luck, and there is no logical basis on which to recommend the algorithms he identifies as having "minimal collisions".