Everytime I read about A/B testing I remember this posting about why you do better than A/B using solutions to the multi-armed bandit problem: http://stevehanov.ca/blog/index.php?id=132
I think "better" needs to be defined. "More accurate results?" Possibly, but probably not by much. Less disturbance to general conversion rates (since the 'successful' case is presented most often barring random variability), sure. But the point is to try out every option, not try out the seemingly more successful option the majority of the time.