If i am the pool that originally mined block 2, then i will get compensated for the lost reward by the double spending transaction, no matter who mines the first attacking block that contains it. In the refined attack, the compensation for the lost rewards is included into the double spending transaction. This compensation is taken from the original amount of 1000 BTC. If someone else mines the first attacking block, i get the compensation from the transaction and the other miner gets the standard block reward and fees and on top of it some fraction of the excessive fee (also taken from the 1000 BTC). He passes on some of the excessive fees to keep the chain more profitable than the original chain. But he got more than he would have gotten on the original chain, because there is no excessive fee available on it.

Graphical version: https://drive.google.com/open?id=0B09kXol8QHGgc2RFMkxjT28xT1E

Maybe the attack needs some refinement. Lets say that the transaction that double spents immediately pays compensation for the rewards of the 6 original blocks to the same addresses. So there is a guarantee that nobody looses rewards in the case that the attacking chain wins. So these rewards can be payed out to miners even if the original 6 blocks get abondened. Furthermore the transaction adds a one time excessive fee. Now you have to decide which chain to extend. The original chain with standard block rewards and fees, or the attacking chain with standard block rewards, but excessive fees. So you extend the attacking chain. Also you pass some of the excessive fees to the next block, because you want all other miners to further extend the attacking chain. Because all lost mining rewards from the original chain were already compensated for on the atacking chain, there is no benefit in mining the original chain. In terms of rewards, both chains have the same length. Miners don't extend the longest chain, because of an arbitrary number, but because their rewards are trapped in it. It is the excessive fee (that must be partially passed to the next block) that makes the attacking chain more profitable. If everybody mines the original chain, the bribe is lost. If everybody mines the attacking chain there is some expected additional profit (and you are guaranteed not to loose your rewards). The collusion is implicit and does not need coordination (assuming all miners want to maximize their profit). No miner looses block rewards. They get the compensation for the lost blocks and profit from the excessive fee.

I still think, there is a connection between security and accumulated miner rewards. Why would you choose not to mine the attacking chain?

But the attack is designed to work without a colluding cartel. It is in the interest (considering mining rewards) of every single miner to extend the attacking chain. If someone extends the original chain, then he does not understand how to maximize his profit. The victim allowed this (100%) attack because of his behavior.

The attack does not need to be announced in advance. Are miners really checking what they are actually doing for their pool? The pool decides what maximizes profit. All they would see is, that they are trying to mine a block with a lower height. So recipients should adjust their behavior now.

Mining pool operators could pay compensation for the lost block rewards on the original chain. So as a miner your expected payout is higher on pools that participate in the attack. And as said above, every pool and every miner has an incentive to participate.

The attack is not designed to be actually carried out. It is designed to question the behavior of current recipients.

Shouldn't we always assume that they are only acting selfishly?

I think miners are already well connected. They can broadcast arbitrary blocks to each other for the others to decide how to react. These blocks are not invalid, they are orphans that arrive few blocks late. There is a possible future where they are part of the longest valid chain. And if there is an orphan that promises higher expected reward than the current chain, we should expect miners to mine on top of it. There is no coordination necessary. Sure, their current software might not be able to identify the expected additional reward. Or it is set up to ignore it. But it might as well accept the bribe.

Miners can minimize the effect on everybody else by inserting all other transactions at the same block height. No other transaction would be lost or even changed. The victim might be someone not well known (unknown?) or with bad reputation. After all, no rule was broken. The blockchain is valid. Code is law. The victim made the error of accepting a transaction as final, before it was safe to do so. Everybody must decide for himself when to accept a transaction as final. No one will come for rescue.

My point is, that the current behavior of accepting a (large) transaction as final after 6 (or any fixed number) of blocks is flawed, because it allows this type of attack. It is not a weakness in the protocol, but in the behavior of its users.

Edit: Modified attack: https://bitcointalk.org/index.php?topic=2115635.msg21171230#msg21171230, Image: graphical version

Lets say i deposit 1000 BTC to an exchange that accepts the deposit after 6 confirmations. I immediately convert and withdraw in another currency. I am totally free to create an alternative transaction chain starting from the original block with 7 transactions that each pay a high fee (lets say 50 BTC) and that can not be mined in a single block, but only in a chain of 7 consecutive blocks (using nLockTime). I then publish this transaction chain to all miners. Even considering the loss of rewards from the original 6 blocks, the expected reward for the miners is higher, when they discard the original chain and start mining a new chain with my transactions.

Obviously my transactions will not be accepted by regular nodes, but i could communicate them to the miners directly.

If a very small miner got one of the original 6 blocks, his expected reward after switching might be less than what he already got. But i can wait for a situation where only large pools mined the original 6 blocks, that have a large enough probability to profit on my excessive fees.

Miners might fight for the blocks that contain my excessive fees. But they can come up with a solution where they share some of the rewards and it is always more profitable to extend the attacking chain.

The exchange might start to offer a fee reward on the original chain. But every reward it pays is a loss. And i am free to pay the full 1000 BTC. In the end I gained nothing, but forced a transfer from the exchange to the miners. They might want to start this kind of attack.

So, how is it save to accept a large transaction (a block that moved thousands of BTC) as final, before the same amount (some fraction?) was paid in miner rewards? Every moved coin is possibly available to bribe miners to construct an alternative chain.