I was referring to the specific attack described in the paper, rewriting history from block one and assigning to yourself all bitcoins, which is clearly a stupid way to steal bitcoins - they instantly become worthless.
I agree, that would be a stupid 'attack', at least not a very profitable one. In the paper it serves the purpose of proving that, even though not profitable, it is possible, and therefore undermining the principle of Bitcoin's block-chain as consensus. At least, as long as not 50% of total existing computer power is used 'in an honest way'.
Regarding merely double spending your bitcoins that's even less of a concern: you still need to amass millions of dollars worth of hardware and millions dollars worth of bitcoins - so that you can double spend them a few times and recover your hardware costs. It also means you need to find a trading partner willing to sell you millions of dollars worth of merchandise for bitcoins, and do so in an anonymous fashion preferably over the internet so as to not get caught. Good luck with that plan.
The temporary mining revenue of 50 BTC/block and later 25 or 12.5 BTC will be worth much more if the bitcoin network is regularly used for multi-million dollar transactions as opposed to buying a few grams of hash or an alpaca sock.
This is all reason why profit-oriented attackers are implausible, or at least their profit will be derived from the failure of bitcoins: speculators, governments, banks etc.
Have a look at the example above, I projected current bitcoin statistics to the moment there's no coin generation anymore. I dare you (or anyone) to alter some input values, like bitcoin value, transaction value, whatever, and I'll try to show such a scheme is still lucrative.
One more question, what do you mean by: "as to not get caught"?