Because performance is not usually a step function. It is a continuum. If it's a continuum, variance in training time to reach expertise is not surprising.