He is saying "If everyone takes the less-travelled road, then to an average magnitude of sqrt(n) where n is the number of previous travelers, a rand()-based cho

He is saying "If everyone takes the less-travelled road, then to an average magnitude of sqrt(n) where n is the number of previous travelers, a rand()-based choice - which can skew the running total slightly - would be an equally effective load-balancing solution versus manually balancing the sides deterministically, while at the same time saving the time spent in examining roads to determine the current level of travel. However, sqrt(n) makes all the difference."