Labor costs are likely double, if not more, than what i've estimated. overtime + insurance + drug testing + security + all the other crap i'm forgetting.

This article: http://money.cnn.com/2011/11/09/smallbusiness/kiva_robots/in... says that 1000 robots costs $15-20 million, not the $500 million that 500k a piece would run. Also what is there to indicate such a hefty need for engineering oversight? 20 engineers constantly overseeing the machines sounds like they are either constantly breaking down, or need a significant amount of custom programming per unit as upkeep, both of which just sounds like a quality control issue on Kiva's side. It makes sense that there's a lot of initial planning/programming for each installation, but I'd expect it to be mostly self-running after that. I would be curious about the energy needs per unit, though. They seem to be carrying around more weight per package shipped than with a human picker carrying just what goes in one package, but then again their use of energy might be more efficient than humans.

> Let's assume that you would need 1000 machines to have as much throughput as 4000 people.

I think your assumption is way off. For one each machine works 3 times as long as a person (24 hours a day vs 8).