When you have the number of visitors (and resources) of Google, it makes sense to actually use A/B/C/D/E testing.

The Google example shows 5 GMail sign-up buttons, but they could just as easily have had 25 variations. When the question is as straight-forward as “what % of visitors to this page sign up”, without caring about the quality of the sign up, then the raw winner should be presented all the time - nothing more to consider.

Of course, if you were measuring things such as likelihood to sign up for other services, become a paying customer, value of purchases, repeat purchases… then simply what gets the most clicks is too simplistic.

A/B testing is usually always cost effective to implement, unless your traffic is really low.