You don't need to solve a block for this to happen, so what is the benefit of having the hashing power?
Not quite. At least today miners who have mempool accepted a transaction will not accept a conflicting one with higher fees, even if they're not attempting to mine the lower fee one yet.
You can pull off doublespends against no-confirm acceptors today, but it's a heck of a lot easier to do it reliably if you have some friendly hashpower.
Well, it's not quite that simple either.
If miners are behaving responsibly, they will have spam filters in place to avoid mining flood/DDoS transactions such as those going to BetCoin Dice.
There are two ways to filter transactions:
The most obvious way is to reject them from the memorypool. This has the benefit of not using up resources upfront. It also means you will accept any conflicting (eg, double-spending) transaction, no matter how much later it appears. For the miner and Bitcoin network, this is all positive. For BetCoin Dice, it means they are very vulnerable.
The other way, would be to accept it to your memorypool, but blacklist it from mining or relaying. This consumes the miner's resources, and if they don't have a sufficiently large memorypool, could cause them to mine fewer legitimate transactions. But it protects the spammers like BetCoin Dice because the double-spend is once again rejected.
When I ran Eligius, I experimented with both solutions at different times, and decided the former (which is better for Bitcoin and the pool, but bad for the flooder) is likely the more reasonable solution.
But to conclude, if miners are acting
in the interests of Bitcoin, it will rationally be rather easy for
anyone to double-spend DDoS transactions.