Flawed Security Economics

I just stumbled upon a piece of economics-of-security reasoning that amazes me:

»Bank robbers steal approximately $100 million per year in the US. (…) To prevent this, banks spend $600 million per year on armored car services and $25 million per year on vault doors. The FBI spends $50 million per year investigating robberies. A good deal more is spent on security guards—approximately 1 million in the US, paid about $24 billion per year (outsourcing makes it difficult to say how many work for banks). In summary, the cost of protecting against bank robberies far exceeds the loss.«

(Michael Lesk, Cybersecurity and Economics, IEEE S&P Nov./Dec. 2011)

I don’t doubt the figures, but the conclusion does not make sense to me. Why should one put the cost of security measures in relation to the losses that they don’t prevent? The $100 million per year are the losses that remain after security. What the security investment prevents is the losses that would occur without it, not the losses that continue to occur despite the effort. I’d love to see an estimation of this quantity. The author even gives a hint towards the possible magnitude, as he continues:

»To look at a less safe society, in a single 2007 bank robbery in Baghdad, robbers made off with $280 million (it was an inside job).«

Perhaps it is even normal for the cost of security to exceed the losses that remain, once the security spending has been optimized?