Yep, agreed. This means that Dropbox's minimum engineering count will be larger than WhatsApp's, but I still maintain that once you have sufficient engineers to tackle the problem, it's still largely independent of user count. Or, put another way, the number of engineers you need is O(problem complexity), not O(user count).
Yes, although I would submit that in a case like Dropbox you get the VC demand for growth which means that if your product is not a natural homerun you are forced to start adding incidental complexity to try to keep the growth engine going. Then before you know it you are O(Investment Money) like Dropbox or Twitter.
Not to take anything away from WhatsApp, I'm sure they have a few lessons Dropbox could have learned from (Erlang more robust than Python at scale?), but it's really not a fair comparison.