There is a great thread over on the Damage Dealing Forums started by a Rogue about the mechanics of the Armor Penetration cap/observed effectiveness. Armor Pen is a statistic that will allow an attack to ignore a given amount of armor.
The thread and the contents inside it are notable in that it is, by my recollection, the first time a Blizzard employee has given out the complete rundown of an in-game formula. While there have been hints and comments about how certain statistics impact the game from patch notes and game designer posts in the past, there has never been a "step a, step b, step c" like algorithmic definition to all those stats contained within the black box of theorycrafting.
In giving out the computations behind Armor Penetration, Ghostcrawler makes note to point out that Blizzard is not, and will not, get in a habit of delivering theorycrafting to players. They like the idea that players have to test out game mechanics, and that while the starting and end results are known, what happens in the middle of combat isn't written in stone. In the case of the Armor Penetration rating, they released it due to quite a bit of (somewhat) inaccurate information out there.
The armor penetration formula, and an example, after the break.