When paying for a (rare) failure is cheaper than paying for the (constant) absence of failure, it's just natural. You know, the optimal amount of fraud in a payment system is not zero. The optimal amount of fatal aircraft incidents is not an exact substitute, bit the pressure is of the same kind, I'm afraid :(