Right now, I think that the way resilience works is broken. The problem that arises, is that the more resilience that you stack, the more each individual point of resilience is worth.
The current iteration of resilience, is that for every 100 or so rating, you gain slightly more than 1% damage reduction. In full PVP gear most classes will have somewhere between 3k and 4k resilience rating, so about 30% or 40% damage reduction.
Lets create a scenario using some arbitrary numbers now.
Class A hits Class B with an attack that always hits for 10k damage. Class B has 100k health, and 0 resilience. It will take class A exactly 10 hits to kill class B.
If we give class B enough resilience (about 1k rating) to grant a 10% damage reduction, now class A will be hitting for only 9k. This means that it will take class A 11.11 attacks to do the same amount of damage. Therefore, we can say that the first 1k resilience provided class B with an extra 1.11 attacks.
Now, if we give class B another 1k resil, for a total of 2k, class A will be hitting for 8k. In order to do the 100k of damage, class A must now hit 12.5 times. This means that second 1000 resilience rating provided protection from 1.39 attacks, a 25% increase.
For a 4k resilience character, probably about average for pvp gear, the 4k resilience will provide 6.66 attacks worth of protection. That means if a toon has 4k resilience, then for each 1000 resilience rating, the benefit was 1.65 attacks. (recall that with only 1k resilience rating, the benefit was only 1.11 extra attacks)
Now you might argue that this critique doesn't factor in stuff like increased damage, crit, mastery etc, but for all those ratings, the reward is either linear (each point is worth the same amount), or diminishing (the value of each individual rating point decreases the higher the total amount). Resilience is the only rating that works in this manner, which will lead to much much longer drawn out battles, and a lot of classes (especially tank classes) stacking extremely high amounts of resilience. Basically, if you increased the attacker's damage by 10% and gave the defender 10% damage reduction, this would favor the defender, and the gap by which it favors the defender would only increase the higher the ratings went. (Consider the limiting case, where attacker A has +100% damage, but defender B has 100% damage reduction).
If anyone's played Warcraft 3, they might be familiar that armor originally worked the same way resilience does right now (each point of armor provided a flat 5% damage reduction), but it was later changed to the more logical approach, where each point of armor provided a 6% increase in the effective hitpoints of a unit. That meant that if a unit had 2 armor, attacks against that unit were treated as if the unit had 112% hitpoints, and attacks were adjusted proportional to that number. The result was that more armor provided the same benefit, but the % damage reduction was decreased per each point according to the formula: (armor)*0.06)/(1+0.06*(armor)). (each point of armor providing a 0.06 improvement in the effective health. In my opinion, this type of mechanic is the only way resilience can avoid this scaling issue, especially in later seasons.