I recall these probabilities from the USCF-Elo system (though they instituted a somewhat different system a few years ago):

67% scoring expectancy for the higher-rated player when the difference is about 130 points (i.e. he would be expected to score 2 points in 3 games).

75% expectancy at 200 points.

90% (actually 90.9) expectancy at 400 points.
99%                                        at 800 points.