That's the one thing that keeps eating at me. Would it not be trivial for a large corporation or government entity to throw enough money around to accomplish this? The only response I hear is that the market will magically prevent this.
But let's just go with your assumption that some entity with big pockets decide to run a majority attack against Bitcoin miners. How much effort and money would it require?
1 year and 20 billion dollars; here is why:
The best 20-28nm mining ASICs require about 1 watt per Ghash/sec. The network is currently at 170,000,000 Ghash/sec so it would require at least 170 megawatt of ASICs, or a ~200 megawatt datacenter (with PUE = 1.15). The fastest TOP500 supercomputer facilities barely reach 20 megawatt. The biggest datacenters in the world are just about 100 megawatt. So the attacker would require two of the world's most powerful datacenters (in terms of electricity and cooling) to attack Bitcoin. But it would take at least 1 year to plan, design, build, and deploy the hardware in two such facilities. And the Bitcoin network will continue to grow during this time. It has grown by 400x in the last year (400,000 to 170,000,000 Ghash/sec). By the lowest estimate, the network will grow at least by 10x in the next year (170,000,000 Ghash/sec to 1,700,000,000 Ghash/sec). So you need to plan to build not 2, but 20 world-class 100 megawatt datacenters by August 2015 to be ready to have a chance to majority-attack Bitcoin. Knowing that a 100 megawatt datacenter costs $1 billion [1], you need pockets with $20 billion in them. And if you under-estimate the growth of the Bitcoin network in the next 12 months, you will have failed and wasted $20 billion for nothing.
[1] http://www.greendatacenternews.org/articles/share/416983/
It would certainly still be feasible for the US government to overpower the network, but it would not be cheap. The Minerscube 15 is due to ship next month, and will get you about 1.666 GH/s/$ [0]. The current speed of the bitcoin network is about 160,000 TH/s [1]. Duplicating this would cost about $96M, assuming you could get that many Minerscube 15 chips at that price, and assuming the infrastructure costs are trivial. So I would say it's still doable, but far from trivial.
[0] https://en.bitcoin.it/wiki/Mining_hardware_comparison#ASIC
[1] http://bitcoin.sipa.be/