In the abstract, what Facebook is observing isn't that surprising. Consider this:
The social network analyzed recent campaigns and found that of the people who showed interest in a mobile Facebook ad in the US, nearly one-third converted on desktop within 28 days. Facebook also found that people who showed interest in a mobile ad before converting were more likely to do so on a different device as time passed.
To me, this is the digital equivalent of the person who liked a coat while window shopping, returned to check it out a few more times, maybe tried it on "for fun" one or twice, and finally, on the third or fourth visit, made the purchase. The behavior isn't new or surprising, but the context is wildly different, and so are the implications.
There's more planning involved in repeatedly checking out a desired item at the store; with a web browser, you have less logistical inconvenience to overcome, and less time to reconsider an impulse purchase (i.e. I doubt that the real world behavior I described results in brick and mortar merchants with a ~33% purchase rate per 28-day cycle, but I'm not surprised that online retailers manage this rate of success).
The other huge contextual difference, as others have pointed out, is that electronic behavior can be tracked much more easily than physical movement. A store salesclerk might become aware of your recurrent interest if you keep going back to admire the theoretical coat-- but a service that can track your browsing will definitely become aware of repeated visits to the coat's Amazon page.
Then again, before long, even of our behavior in physical stores might be mined for data. Your smartphone (or smartwatch, or connected eyewear, or whatever) will know when you're in proximity to stores, and when you enter them. It will be able to report to a "beacon" (or whatever they end up being generically called) where you linger, what objects are nearby, how often you return to certain locations, and, finally, when you buy. I think that's a natural extrapolation of some of the stuff Cisco and Apple have already talked about, for example, and an example of the way data will generate targeted ads, hopefully without feeling too much like Minority Report. That kind of service will (or at least should) require that users opt in for personalized ads, which is a bit different than the anonymized data described in this article. Still, I don't doubt that anonymized location data will be used similarly.