What I don't understand is how you would apply A/B testing to SEO. There's only one Google; you can't show half of Google one version and half another Google. How do you get data on which version is better for search engines without being able to compare them while holding other factors constant?
You can try to simulate a 'test' with two equal throwaway sites, but this will never give you the same insight as using it on a big site with real content and real visitors. But it can help you find problems in linking structures, tweak the (really small) benefits of using one or another html tag.
The success of SEO can't be isolated. But, increased crawling of the page and disproportionate visitor growth from Google are good signs.