Doesn’t multi-variate testing using DOE also accomplish this? In fact, DOE also exposes how much each variable affects an outcome, and how a combination of variables interact.

http://www.isixsigma.com/dictionary/Design_of_Experiments_-_DOE-41.htm

The problem I tend to run into with pure A/B testing is that it’s very dependent on the order in which you conduct your tests. For example, testing having a button on the left or right of a landing page as well as the text and design of the button. If you test left vs right first, and left wins, you’ll never know if having it on the right with alternate text or shape might be even better because you would have assumed that left is your winner. This makes it very hard to come away with any concrete findings. But yeah, DOE’s are harder to do, and usually require more complex software.

But the good thing is that, once you actually break down the drivers to its sub-components, there are probably not more than 5 pages/features that drives the overall goal that you are trying to achieve.

And I agree with you a 100% on this:
“Recalibrating the “core mechanic” of the product to make uploading photos a natural part of using the product (like HotOrNot, for example)”

This is where we got the biggest bang for our buck…