B
Did the same test with my shadowblade / shaman combo. First test was done without buffs. My blade has 182 dex / 136 quick and dodger 2.
100 swings (hit or evade)
48 evades
52 hits
Second test was with a dex buff increasing it to 244.
52 evades
48 hits
The first test has a margin of error of (.292)(.708)/sqrt(264-1) => +/- 1.27% per confidence interval. Three confidence intervals is 99% certainty the actual evade chance is within +/- of 1.27% x 3, or, 29.2% +/- 3.81% for this scenario.
The second test has a margin of error of (.276)(.724)/sqrt(272-1) => +/- 1.22% per confidence interval. Three confidence intervals is 99% certainty the actual evade chance is within +/- 1.22% x 3, or 27.6% +/- 3.66% for this scenario.
That means the two probability curves overlap by a large amount. If we go down to two confidence intervals, they still overlap by a good amount. If we go all the way down to one confidence interval, we still have overlapping probability curves.
If DEX has an effect upon Evade, it's damn hard to detect. For all practical purposes, it's small if it exists at all.