There were indications during the presentation that it actually runs face recognition often, including as a power-saving technique (sounded like it would turn screen off when you weren't looking at it, back on when you do).
I'm not a camera hardware expert, but my understanding is that (like most things), the booting process uses up more power than the same duration of time in a steady state. So the "five minutes" number is misleading; if that 5 minutes is broken up into hundreds of discrete startup cycles (which is not at all unrealistic - people instinctively check their phone a lot more often than they think – and there's no way for the phone to tell whether you're turning it on to glance at the clock or to unlock it until after it's turned the camera on), then it's a significant amount of power draw.
