During the experiment of not displaying points on comments, was the voting less volatile, with smaller voting swings? If it was, then hiding the points until a few minutes or hours after the comment is posted (i.e. until the volatility becomes reasonable) could replicate the effect. If not, then maybe the voting swings are inherent as a statistical property of any sufficiently large community, in which case you may want to display a confidence interval of the voting score instead of the score itself.