I understand what you're saying and think it is plausible given your assumptions. For example, you think that Dropbox's primary acquisition engine is viral referrals, that the number of referrals it gets over a certain interval is the number of new users times some constant conversion rate, and that you can't just increase that number of referrals by dumping money at it like you could increase the number of new users by dumping money on an effective CPC campaign.
I think your assumptions are likely to be contrary to material fact. In particular, I think that the embedded assumption of "constant conversion rates" is false. For example, I think that inducing a customer to give a referral is a conversion like any other, and that one could spend a non-trivial amount of engineering resources to instrument one's site and then spend engineering/marketing resources to A/B test the "#$#$& out of it, and this process would predictably result in increased conversion rate. (Test the call to action. Test the design of the form. Test integration directly with Facebook. Test how much free space you offer for a signup.)
Do I have to mention that A/B testing in the context of a viral loop is hypereffective because of compounding? You get exponential returns to linear improvements. This is on top of A/B testing tending to give exponential improvements to linear work, because today's improvements compound on yesterday's.
The combination of these two factors means... wow. That's like a repeatable recipe to... wow. Crimety. I think I have to spend some more time thinking about this later, but my snap impression is that you could have Zynga-esque growth in an arbitrary niche (i.e. selling to people who pay money for things) without scamming anyone.