You forgot to add the key qualifier "at Google". That says nothing about how being good at programming competitions correlates with being good on the job in general.

It's also kind of common sense that if you spend a lot of time working on something that's not really related to your industry job that won't be as good at said industry job as if you spent that time, say, writing open source machine learning software. People just assume that programming == software engineering, but programming competitions are a lot more similar to math competitions than they are to real world jobs.

Norvigs claim can quite probably be credited to selection bias. See http://erikbern.com/2015/04/07/norvigs-claim-that-programmin... for some nice graphs.