Dr. Awkward said:
Two things:
People who do probability distributions call every continuous function a curve, even when that function is a straight line. Quite often, statisticians attempt to fit data to a linear curve because it illustrates a particular type of correlation.
The scale of the difference is important. If you differ by 6 points, it doesn't matter whether you differ between 3 and 9, or between 9 and 15. Your relative abilities will be the same. Compare this to a hypothetical system in which your bonus is the square of your ability. In that case, the guy with a score of 3 would be rolling +9, and the guy with the score of 9 would be rolling +81, a difference of 72. The 9 and 15 pair would be rolling +81 and +225, respectively, a difference of 144, twice the magnitude of difference. In a linear system, a given difference between ability scores is always worth the same amount. In a non-linear system, a given difference is worth different amounts depending on the values of the particular ability scores.
In a linear system, having a 10 Str compared to a 12 Str is exactly the same as having a 18 Str compared to a 20 Str. Your opponent always has a +2 advantage.