It would be interesting to know how much data Google stores in traditional RDBMS versus NoSQL stores (BigTable, Spanner[0], etc.). I could probably track down this number, but then I expect I wouldn't be able to talk about it after, so for now I'm saying ignorance is bliss :). I would wager that we're rather more NoSQL heavy than most organizations, so it wouldn't really be a good proxy for the industry in general, but it might be helpful for justifying Steve's claim. It might be particularly helpful if we looked at the workloads Google runs on NoSQL stores that traditionally would have mandated an RDBMS.
[0]: http://research.google.com/archive/spanner.html