The technology isn't here yet; that's the problem. Book printing machines that can do perfect binding and colour covers are, as I understand it, ferociously expensive to build and run: you're talking $50K and up for the machine. Next, the product they produce isn't quite as good -- the closer you want it to approximate a traditionally printed and bound book, the more expensive the machine (and the printing process) gets. Finally, the running costs for such machines are much higher than for a printing press geared up to run off a thousand or so copies of a single book -- up to an order of magnitude higher. When you buy a paperback, maybe 10-20% of the cover price goes on materials; if you buy a one-off POD book printed this way you're looking at 60-70% of the cover price being printing costs and materials.

This wouldn't be a killer, except that you're adding a delay to the bookstore browser's experience, a delay of several minutes before they can actually see the book, which is usually a necessary prerequisite to buying it. Someone who's in a hurry isn't going to want to wait -- especially at busy times of day when there's a queue. Someone who isn't sure they want a specific title will be loath to order a copy printed. And so on.

I think something like what you describe will happen eventually, but like the perfect ebook reader it's quite a way away. And it's going to be more like coffee shops acquiring book printers (to go with the free customer-attracting WiFi) than bookstores changing their model. I reckon they'll use ebooks as a promotional tool: pay $5 for the ebook, which is steganographically stamped with a rebate code. Send the ebook and the rebate code to the printer along with another $10, and collect a hardcopy to take away, where the hardcopy actually costs $14 if you buy it separately. Something like that might work.