The only problems I see with the setup Bob's describing is that RFID tags don't "go off", they just respond to interrogation. So, the tags wouldn't be able to tell you if someone just ducked down and crammed a twinkie down his throat, you'd need a camera for that. Also, it'd be impractical to interrogate every last RFID tag every minute or so; you'd need to have a unique ID for each and every item in the store, cross-referenced with your inventory. If there's ever an inventory screw-up, or if an item fell behind a shelf and got blocked from the scans, your entire store would go into lockdown.

You're better off with the basic camera setup. Instead of relying on radio waves, you can use your eyes to see what's going on. You can vary this a bit, and instead have a score of fixed-drones acting as your cameras; this gives them a lot more in the ability to raise a practical response.

Actually, given the amount of processing power available in 2070 (just look at how AR works... constant bombardment by actively transmitting devices is considered the norm), I wouldn't really have a problem with a system based around querying all the present RFID tags once every couple of seconds. In a Stuffer Shack, that'd only be a couple thousand objects to keep track of, tops. It'd barely show up as a percentage of processor time usage. This doesn't mean that it has high-end decision making ability... if a twinkie moves out of its designated area, the system is going to assume someone was trying to buy it, request cash from their commlink, and set off a simple alert if the cash is not forthcoming. edit: The attendant glances at the camera, sees if he's got an armed incursion of hungry customers scarfing food off the shelves, and makes the lockdown decision himself.