No, this is completely wrong.
If you have majority hashing power, this allows you to
a) entitle yourself to a greater proportion of the newly minted bitcoins (though you actually can not increase the global rate of creation), and
b) undo already "confirmed" transactions, within a small timeframe. This allows for double-spend attacks, but you can only do that if you owned the coins in the first place. Nothing, not even hashing power, lets you use other people's coins.
In particular, you can't "rewrite history" arbitrarily, due to hardcoded hashes in the Bitcoin clients. You also can't "make all the bitcoin ever mined owned by you".
Edit: Let me clarify, to actually steal coins from someone, you would have to forge a transaction, which takes the same effort as mining say an SSL certificate for google.com.
Instead of selling the machines for $30k a piece, could they not just use the machines themselves to transfer all bitcoins in the world to their own wallet? The total amount of bitcoins mined to this date should value much more than 12*$30k?
Does that mean that the entire bitcoin network should as of today be considered compromised, game over, pack up and go home?
The inventor of Bitcoin once described this scenario, saying it would be smarter and more profitable to simply mine bitcoins legitimately...
A government concerned about losing control of their currency is not going to behave the same way as an independent actor looking to make some cash.
I'm not saying it's going to happen; just that the risk should not be discounted.
A double spend attack only allows an attacker to double his BTC before the block chain has to be forked, and before the whole community notices the attack. You can't send the same BTC to multiple exchanges. You have to execute one double spend attack for each exchange. But the community would detect the attack after the first one.
(Double spend attacks would be noticed when multiple blocks in a row are replaced or, in Bitcoin's terminology, when a reorganization occurs: http://blockexplorer.com/q/reorglog)