"... most people are going to have to slog away somewhat tedious things like making context ads match their page content 0.1% better."
I'm not sure why you think that the 0.1% improvments are assumed to be a tedious slog. They may have required some seriously interesting new ML techniques just to get that extra boost. Recall the Netflix recommendation algorithm improvement contest that had teams working very hard to get those last fractions of a percent.
Small adjustments to a huge system can have huge impact: http://norvig.com/speech.html
> What makes Google what we see is that relentless pursuit of increasingly tiny increases in efficiency.
Absolutely. This fact is literally in the comapany's dna. A lot of issues that were mentioned in that article are a direct result of this.
I'd be curious to know how this affects all of the startups they acquire that remain dormant for so long.
You also tend not to get a full view of things at Google. If you're in backend engineering, that's what you work on. I know that I like seeing the full product and knowing that I've worked on what the users are going to see.