Separating the requests allows for different cache policies. If you simply delegate the entire recursive resolution work to Google (or whomever), they get to record that you needed "www.example.com" every TTL (5min?). You don't even need to see the domain's NS records in that case, so there isn't an opportunity to choose a cache policy.
