You listed a lot of dimensions, whereas to make the correlation between mac address and customer info, you only have 1 dimension (visits/location) to do the correlation with.
In practice, I find that I've needed a lot fewer dimensions than I'd expect to find people; when I was in grade school one time I cross-referenced the attendance list (for a last name), the reverse phone lookup in the white pages, and a map of the school district boundaries to figure out where a classmate lived.
I think that repeated visits to a grocery store like a Safeway would be a really good proxy for home ZIP code. Every Safeway I've visited in the SF bay area has a wifi access point, and I wouldn't be surprised if it was logging the MAC addresses of phones that pass by. Maybe some hackers will try to optimize their route home and pick a grocery store that's half way between work and home, but I hypothesize that a large enough number of people choose to run to the nearest grocery store for that "one item they forgot to buy for dinner" or for dedicated weekly shopping trips that you could draw reasonable conclusions about aggregate behaviour.
[0] I like buying stuff online; it's kind of unavoidable there.
[1] I always get this weird sense of power when I buy peaches at the Safeway at full price during off-peak and then the next week they go on sale for "Safeway Club" cardholders at the store I usually shop at.
For example, imagine they have cameras that can image your license plate. You go there twice -- they have one MAC and two sets of possible plates. The odds that you and another person were both shopping at those times is pretty low. Now they have license plate, make/model of the car, can probably triangulate the wi-fi to know what you bought each time with reasonable fidelity...