Meanwhile I have come to the conclusion that it can be easily proved that as soon as bitcoins would become truly valuable, it would be lucrative to fraud the system by gaining more than 50% of that cpu-power.
When they become truly valuable, transaction fees will be worth far more than they are now, which will increase difficulty, meaning the cost of attaining 50%+ of the hashing power will be far higher than it is now.
With 'truly valuable' I mean value that can be depended on, no matter what that value is. I certainly don't want to speculate about whether that value be higher or lower, but if it were higher, transaction fees would be worth more than now. But if I do a rough guess of the transaction fees (by inspecting a few blocks on the blockexplorer), they're now about 0.05 - 0.20 bitcoins per block.
That means, if exchange rates wouldn't change, a ROI of 0.2% of what it is now and with that an expected difficulty of 0.2% of what it is now. A not so huge investment is necessary for that.
I know, fees could rise, the exchange rates could rise, the number of transactions per block could rise. However, it's always safe to assume fees will be much less than total transaction value in a block, and therefore it's lucrative to calculate hashes of an forked block-chain with double-spent transactions.