DEX doesn't increase accuracy.
This would be difficult to test, but doesn't necessarily require a massive sample size. We know that damage varies within a fixed range. We can use that knowledge to compare parry damage taken to unparried damage taken. You'd still want a decent sample size per test point to be sure, true, but we're talking 500-1000 samples instead of 50000. Such a test is not without complications of its own however, as you'd likely run into resolution issues on monsters that deal low enough damage to be otherwise ideal test fodder. Such a test would require the use of a monster that hits fairly hard, ideally at least several hundred damage per unparried hit. You might need a second individual, you'd have to pay more attention, and rate of data gathering is slowed.