I think you have a valid point. Let's say I control 50% of the mining. How much would I accept to start a new fork from 10 blocks back?
With 50% of the mining, I'd be getting about half of the blocks, or 5 of the 10 blocks just mined. Asking me to fork from 10 blocks back would mean I'd be giving up these 5 blocks worth of payment.
The question then is how long it would take me to catch up to the real block chain from 10 blocks behind. At 50% of mining, I'd be flipping a coin to try to get 10 more heads than tails. On average, that takes some number of flips, say N. So, I would have to mine an additional N blocks, on average, to catch up to the current chain.
So, if you pay me more than N+5 blocks worth, then it would on average be more profitable for me to start a chain for you than to continue the current chain.
However, this is the average, and the distribution has a long tail, meaning it might take hundreds of blocks until the new chain is longer than the current one. There are some ways we could deal with this risk:
1) I charge you a per-block fee, and you pay me as long as you want me to mine for you. There is no promise I will ever catch up to the current chain, but I'll keep trying as long as you keep paying more than I'd make with legitimate mining.
2) I charge you up front and take the risk. In this case I'm on the hook to catch you up to the existing chain. I'd probably charge a large premium based on the potential to divert my mining resources for a long time. Also, it's possible I never will catch up, so I'd probably add a refund clause to cut my losses if it looks hopeless.
The idea that I'd mine legitimately because I want Bitcoin to succeed has some truth, because there is some setup cost to mining and I want a return on this investment, but for the right price I'd probably take the risk of killing Bitcoin. That risk is captured in the premium I charge you over my normal mining income.
Plugging in some numbers, I'll use Briguy37's assumption of 15 BTC per block mined, and let's assume N is 20 and my premium is 100%. Then, I'd charge 25*15*2 = 750 BTC to do this fork.
We can argue the exact numbers, but the main point seems to be what you were saying, which is that large transactions are not as trustworthy as smaller ones.
I would take this one step farther and argue that some smaller transactions may not be trustworthy either. Suppose I agree to pay 100 people 10 BTC each. That's 1000 BTC from me, but each transaction is only 10 BTC as far as the recipients know. if I time them all the same, the cost to reverse all of them is the same as if they were a single transaction.