Typically, when economists speak about optimizing something, they think about maximizing the surplus of benefits over costs. "Cost - benefit ratio" is a popular word, but the actual ratio between benefits and costs (the marginal rate of return) is not what you should optimise.
For example, an investment of 1 for a benefit of 10 (ratio 10, surplus 9) is a much inferior outcome to an investment of 100 for a benefit of 200 (ratio 2, surplus 100). It's thus better to speak and write about cost-benefit surpluses rather than about cost/benefit ratios. The latter is much more common and even I do it often, though.
I've looked at military bureaucracies and their products - both historical and modern - for decades, and I can tell you they do not usually optimise either ratio or surplus of costs and benefits. They are not being restrained in a way that would lead them to such an outcome. Instead, all things military appear to gravitate towards another theoretical fix point: The limit of tolerance.
This is a different way of describing what I wrote in August 2013-08 Niskanen's bureaucrats, of course. The "limit of tolerance" is a more general expression than to say "they spend everything they get".
Military bureaucracies do not call for an optimised budget, but for the maximum budget the constituents would tolerate. They do not equip the infantryman with an optimised load, but with the maximum he can carry (or they can convince others that he can carry). Vehicles and platforms aren't going to be optimised in their equipment, but maximised to the limit of the budget-makers. They don't re-organise when it's wise, but when the old organisation becomes unsustainable. They don't adopt unpopular novelties when they're worth it, but when backwardness is no more tolerated. Poorly-suited leaders will not be removed once their unsuitability is discovered, but once its effects become intolerable. Force structures are not optimised for military effectiveness, but have as much of the fancy officer posts (HQ and command slots, fast jet pilot seats) as possible while providing as much of the undesired components as necessary in order to achieve the master's tolerance. Air space deconfliction doesn't strive for minimised overall friendly casualties; it covers the decision-maker's asses as much as can be tolerated without a collapse of the military effort.
Maybe the insight that the actual outcomes are the product of the full exploitation of tolerance rather than an optimisation of benefits surpluses can help us to determine how the actual optimum differs from the status quo, and how to correct the normal outcome in order to get there.