Indeed. If the razor has to be applied I'd go with A because it's simple and feasible enough solution for a company like Google and because B would imply virtually zero levels of corporate worries about the eventual PR downfall. The fact that It's Arrington we're talking about here and not your next-door blogger, reinforces the probability of A being the simpler choice.

No way. It would require poring through so much more raw data to take raw tcp traffic, structure it somehow and pull out email messages (option a). Compare that against going through structured data in the gmail storage system (option b). Option B is much, much simpler.