If two out of three can be wrong, why can't two out of two be wrong? If you have a fail-safe, you should always do it if even one sensor is off, regardless of the majority
But if you don't have a fail-safe, or if the fail-safe is too costly, or if the sensors are not that reliable, with 3 sensors you have more chances to make the right decision by following the two sensors that agree. With with only two sensors, you'll have to follow one of them, i.e. you have a 50-50 chance of making the wrong choice.