Challenging the Mean Time to Failure: Measuring Dependability as a Mean Failure Cost