It's not that hard. Basically they're just taking a diff (binary in this case) between the new version and the old version, uploading the diff, and applying it to the version stored on their servers.
You could very well be right about what they're planning but if that's the way they're going it will end up killing them in the end. As someone who administers Dropbox in a team environment the biggest issue is its creation of "Copy" files whenever there are simultaneous edits.
I mean, think about it. If you're using Photoshop and your edits go into a copy file you basically have to re-do that work. There's no easy way to merge changes (I guess if you were working on different layers but still it's difficult).
And (as I said below and got downvoted for) it isn't that hard. We use Dropbox for Teams and I back it up with a custom application using components from a company named Aspose. They allow me to read Office file formats and combine files where needed.
If I can do that in a weekend then why shouldn't Dropbox look into doing it?
In source control systems like Subversion the application can usually automatically merge changes that happen in different parts of the document, but will require user intervention to merge conflicts. Merge conflicts can be difficult to resolve even in plain text. I really don't know how you could automate that in plain text files, let alone arbitrary file formats.
You use Dropbox's already existing functionality to make a copy file. You then compare the changes and if they don't conflict you merge the files (while backing up both just in case of error). If they do conflict you send a message to the tray application which displays a popup showing the conflicts and asking for a choice. This isn't rocket science.
There are certainly places where that wouldn't be feasible like video editing. But I don't buy the argument that "because you can't do everything you should do nothing". Again I point out this took me a weekend to do with office documents and I didn't work through the night to accomplish it. So you can't really tell me it's out of the grasp of a company that has millions in funding.
There are also cryptographically signed documents which you cannot merge, because that would break signatures.
So the question is, how much automated merge processing should be done, and how big are the risks in doing so. Even if the system were capable of doing merges for some documents, it would need to have a setting to turn it off. Recovering erroneously merged documents also needs some thinking. And so forth.
So bottom line is I guess we agree that in some cases merges can be done, automated to some level even, but there is a lot of work needed to make it safe, understandable and usable for the majority of users.