Bitcoin Gold: https://cointelegraph.com/news/bitcoin-gold-blockchain-hit-b...
Vertcoin: https://www.coindesk.com/the-vertcoin-cryptocurrency-just-go...
Ethereum Classic: https://cointelegraph.com/news/ethereum-classic-51-attack-th...
Google is happy to show you more.
A 51% attack only lets people steal money by taking back recent transactions and sending them elsewhere. Transactions that have been received a long time ago are generally safe. (An attacker can reach further back in time if they run the 51% attack longer or with even more hashpower.)
51% lets you rewrite arbitrarily-deep history, limited only by how much time you're willing to spend on the attack. You have more hash strength than the rest of the network combined! You pick some ancient block and start building blocks off of it, and the very definition of a 51% attack is that your new fork will eventually outgrow the "real" fork.
Double spends are possible well before 51%. Even a 10% attacker has the ability to (sometimes) double-spend. A 10% attacker can try to double-spend transactions and has a certain chance of being able to. The chance of success decreases exponentially, the deeper in the chain the target transaction is. However, for a high-enough value transaction, attempting to double spend it can have a higher EV than the mining revenue you lose out on by mining honestly.
The potential losses are huge (especially for exchanges etc.) from double spend attacks, and the value of a cryptocurrency is very much dependent on users' confidence in its resistance to attacks.
If it's not profitable, it is unlikely to happen. If Hashrate was to drop significantly without commensurate price drops, attacks will happen and then the price will drop to compensate.
However I am confident based on previous events that BCH prices will trend strongly downward when news of the next successful 51% attack is released.
However, if the network is unstable, then that balance will lose value in real terms or even become worthless.