It's not an option you just turn on that does things. It is a place for you to access the physic's state of a RigidBody and write your own integration based on that information. So you can get the collision results, but then you will write how it should handle that.
Without any working knowledge of your problem, I suspect it could be based on the speed of the rotation and the collision detection method being used. Maybe try setting it to continuous collision detection. I haven't looked very far into the physics code, so I don't know what they're using, maybe it will switch to some form of conservative advancement. That's purely speculative on my part though.
I've only done a small amount of reading into this area, but it doesn't appear as if there are any perfect solutions, at least nothing optimal for games. Every method has some sort of trade off or constraints (performance vs. accuracy), and you as the developer must know them and decide what best fits your game, or how you can design around them. (IE - Sometimes thicker walls, or extra colliders that give hints are what is needed for a particular problem.)