Somebody said earlier that there is a hidden probability trap - we want to treat non-random information as if it were random.
I see the simple answer as hinging on a probability function that collapses when the warden tells the prisoner the bad news about his buddy in cell C. This is the similar to that phenomenon called quantum entanglement and is a good illustration of the idea that state of a certain quantity is a function of certain constraints, and the function returns a different answer as the constraints change. A Bayesian solution based on conditional probability analysis looks at it from an entirely different point of view and yields the same answer. Which is also very cool.