I'm curious as to the long-term economics of mining.  I'm interested at what happens when bitcoin gains price stability, the majority of bitcoins have already been mined and (presumably) miners mine to win transaction fees rather than new bitcoins.

Suppose all miners are rational economic players, in that they mine if and only if the rewards outweigh the costs.  This means that it is in effect free to buy hardware and electricity to mine with - every GPU and kWh funds itself by winning transaction fees.  This in turn means that (with access to large amounts of initial capital) you can actually build an arbitrarily large self-funding mining rig.  You just keep adding extra nodes/GPUs because each node is self-funding.  Eventually, by accumulating more nodes you end up owning half the network at which point you can (it seems) do many devastating attacks on the network.  Given the time you've invested you may not want to destroy it completely, but you may wish to make some huge double spends that live for long enough for you to cash out into USD.

It seems to me that the only way of stopping people taking control of the network for free, is by making mining being a loss-making activity...but then why would people want to mine for free?  But even then, you would still have to make it significantly loss-making to stop black-hats from 'investing' some short term capital with the reward that they can do some monster double spends.  So you can see the general argument here - it costs the same amount for a good-guy to mine as it does a bad-guy, except that the bad-guy has the added incentive of double-spend upon seizure of the network.  Economically speaking, this means that in the steady-state era it is rational to mine only if you're a bad-guy looking to seize control of the network.

Hopefully I'm wrong about all this and I've missed some key argument...but what is it?