The important part is that the authors (and several others) have verified the statistics out to a few hundred billion primes -- and that while the bias does start to drop out, it does so "very slowly." That's what makes this result not "numerology."