I've given some thoughts about what nasty things an evil miner could do, and I think I found two problems that I'd like to share:

First of all, a miner is supposed to follow the following rules when including transactions (ordered by priority in descending order):
1) In case of a double-spending attack, the oldest tx is the legitimate one. Only include the legitimate tx in the block.
2) Include as many tx in the block as you can.
3) Maximize your profits by prefering tx with higher fees over tx with lower fees.

And here's the problem. Miners should follow rules 1 and 2 even if they could make more money by braking them. I do think that eventually, miners will show up who have different priorities.

First, 0-confirmation double spendings aren't new. However, I'm afraid that eventually, some rogue miners will show up and create a market for double-spending attacks. I mean, solo miners who try to put wrong transactions into the blockchain are one thing. Now imagine what would happen if several rogue miners got together and exchanged wrong transactions with each other. It would combine their hashrate and increase their profits because now, they wouldn't have to mine blocks for their rogue confirmations themselves, they can also buy wrong confirmations from other miners. This would be dangerous for the network as a whole because it would decrease the costs of confirming a non-legitimate transaction.

The second issue is that miners could artificially slow down transaction confirmation. The confirmation process is supposed to be a competitive market. Supply is limited by the maximum block size; Miners will always prefer those transactions which include the highest tx fees, however they will also try to include as many tx as possible in order to maximize their profits. However - there's a weakness in the system: Miners can reduce supply artificially by mining smaller blocks. This way, users would have to pay higher transaction fees in order to have their transactions cleared. A small transaction fee increase may appear acceptable. And we might hope that this guy doesn't have enough market power to reduce network-wide supply to the point where it increases his profits. However, there are cases where a user doesn't want to wait for the block after the block after the next block that will finally include his transaction; Instead, the user may want the transaction to clear NOW. And now, let's look where we can buy "transaction inclusion in the next block". We can buy that service only from exactly ONE miner. We don't know who's going to mine the next block before it's happened and each time it's someone else. Yet, there is always only one miner who is going to mine the next block. This situation is called a monopoly. Rogue miners could exploit it - they could require excessive transaction fees from those users who want to get their tx confirmed ASAP. When a user refuses to pay the inflated tx fee, the tx would have to wait until a sensible miner creates a block.

The combination of both scenarios is another problem. If there are too many miners around who refuse to include legitimate tx we have a higher chance that a non-legitimate tx might make it into the blockchain because now there are SEVERAL miners who have the opportunity to include wrong transactions.

Bottom line: At any time, there is only one miner who is going to create the next block. This miner is a central authority which has the (temporary) monopoly on tx inclusion and the authority to make almost-final decisions about double-spending transactions. Even though the powers of this central authority are passed around all the time, it is still an issue in the above scenarios.

As a solution, I think that we need double-spending safeguards that relieve users from the need to rely on the moral integrity of a particular miner.