What happens if I play the same player many times?
The ratings algorithm uses many reference points to build connections and make relative comparisons between the universe of players. By only playing against one other person, the algorithm does not have as much information to make an accurate determination of a player's general level of play.
For instance, if two players only play against each other and are 50-50 in their results, all that is really known is that the players are roughly equal. The question still remains where those two players fit relative to other rated players.
Crossover in match results between different players is what allows the entire country (and even international players) to calibrate on a unified scale (i.e. a 3.5 in San Francisco is the same as a 3.5 in New York). As players play a diversified set of players, their individual ratings will improve in accuracy.