The cost of a 51% attack has been estimated to be $6–$20 billion [1]. This is only 1.3%–4.4% of Ethereum's current market cap. And it is only 0.5%–1.5% of its potential growth if Ethereum knocks down Bitcoin and conquers its full share of the cryptocurrency market.
6-20M to reverse how many blocks exactly? The way PoW deterrence works is that the cost is disproportionately higher than the result of such attack. Spending such high percentage of total market cap would be viewed as a loss for Ethereum, because it means that after a couple dozens of such attacks Ethereum's value will collapse, will Bitcoin will just suffer from a temporary set back. And the practical damage of 51% attack can be mitigated by requiring more confirmations.
6–20 billion, so hundreds of blocks, really.
No, the potential gains of a 51% attack on Bitcoin is much higher than the operational costs (since the transaction volume is much greater than the mining costs). And especially if we are assuming that the value of Bitcoin will recover afterwards, as you do. Then the attackers would just be able to keep coming back for more, once they have gained a majority.