I don't want to repeat myself. You will destroy the security of the coin by enabling the hashrate attacker to steal other people's coins.
You mean that you could steal all trades that sell a particular altcoin by rewinding that altcoin? That means that the altcoin probably has low hash protection.
It is a good point though that rewind risk does need to be incorporated into the system. The current implementation has steps that are zero-confirm, you wouldn't want to do that with high value trades. It is slower, but worth it due to the higher risk.
A rule of thumb is that 1 block "costs" whatever the block reward is (inc fees).
At the moment, it costs 25 BTC to mine a bitcoin block.
You could set the minimum confirms to k * (trade_value) / (25 BTC). The k parameter is some protection against multiple trades being attacked together.
With a k of 10, a 2.5BTC trade would have a min confirms of 1 block. A 100 BTC trade would have a requirement of 40.
If the altcoin pays out 0.5BTC of value per block, then 50 confirms would be required on the altcoin.
This is an extra requirement. There would be a minimum number of confirms, no matter how small the trade.
Since the fee can be done on bitcoin, which has strong protection, these confirms would be mostly for the altcoin confirm.
Another attack would be to stall the altcoin chain. You could make blocks and push up the difficulty so it stalls.
I think these attacks mean that the altcoin is weakly protected already. Even a centralised exchange could have problems. They pay you in altcoins and the transaction gets rewound. Having said that, they would probably refund the money.
They also do the "lots of confirms" thing. For weak altcoins, exchanges require deposits to have lots of confirms.