I at least know someone who worked (tangentially) on the statistics thing, and it was in the pipeline for a long time. Similarly for things like searchwiki or longer snippets (I've heard of small UI changes sitting around for years).
I think whats really happening is that when competitors do things its often a validation for the team that is doing it at Google. Further, I think some of this is kind of silly because this is Google we're talking about: the guys who sat on GrandCentral for well over a year in radio silence to... err... migrate it somewhere? Google development methodology is "do it once, do it right", and by God they're going to work on it until its right. This isn't some market analyst telling the engineer to turn up the "copy Cuil" knob or anything.
From the article: "Can you imagine Walmart making wholesale changes to its stores because mom and pop's store on the corner implemented some neat features for its customers?"
Yes, if Wal-Mart had the magical power to allow 1% of its visitors to walk through the door to a completely different store, they would have the power to experiment with a lot of things that would arguably make the shopping experience better.
I have no doubt the Google A/B tests a lot of this stuff before rolling it out.
My understanding is that major walk-in retailers like Wal-Mart do multivariate testing to maximize the conclusions they can draw from simultaneous experiments. Apparently an approach called 'taguchi split testing' is especially valuable for evaluating combinations of changes, and is also applied in online situations that go beyond simple A/B testing.
I think whats really happening is that when competitors do things its often a validation for the team that is doing it at Google. Further, I think some of this is kind of silly because this is Google we're talking about: the guys who sat on GrandCentral for well over a year in radio silence to... err... migrate it somewhere? Google development methodology is "do it once, do it right", and by God they're going to work on it until its right. This isn't some market analyst telling the engineer to turn up the "copy Cuil" knob or anything.