A few years ago, I saw a Hybrid PoS/Work chain (a satoshi-client derivative coin) being attacked in real time, and totally hijacked. Ultimately it was a bug in the PoS logic that enabled it. I am very apprehensive about PoS even providing a fragment of the security. Chain takeovers are a real threat in PoS world. The attacker needs very little physical resources -- just know the right stakeholders. PoW need resources. PoW is trivial to verify and test. PoS advocates like to gloss over all the real world failures.
You may notice that during the China ban, Bitcoin worked pretty much as expected (with some slight block delays until difficulty adjusted). PoW works.
The other metric to observe is how many chain reorg's occur due to PoS. This is practically a non-event in large PoW chains like Bitcoin, today. In Turing complete chains like EVM chains, it might not be possible to reorder transactions with no-loss during reorgs as transactions are state dependent. PoS potentially destroys finality.
Do you mean [1]? It sounds like the consensus issues there were related to Solana's unusual PoH model. Validators being "unable to process all the proposed forks" would never happen in a BFT PoS model, like Tendermint etc, since forks cannot occur (under a 2/3 honesty assumption).
> Ultimately it was a bug in the PoS logic that enabled it.
To state the obvious, PoW chains can have bugs too. It sounds like you were just looking at a niche project that didn't have high standards for engineering and audits.
> PoS advocates like to gloss over all the real world failures.
PoS and PoW chains have both had plenty of incidents, but you haven't really pointed out any PoS incidents that were caused by weaknesses inherent in the PoS security model.
> Bitcoin worked pretty much as expected
Bitcoin has been around the longest, has a relatively small feature set, and rarely changes. One would expect such a chain to be very stable regardless of PoW/PoS.
> The other metric to observe is how many chain reorg's occur due to PoS.
Reorgs can't occur in BFT PoS chains (under a 2/3 honesty assumption).
[1] https://solana.com/news/9-14-network-outage-initial-overview
Contrasting: https://etherscan.io/blocks_forked and https://polygonscan.com/blocks_forked it would appear that indeed PoS BFT chains like Polygon can and do reorg, very regularly.
At this point PoS is quite prevalent among modern blockchains (Cosmos, Solana, NEAR, Algorand, Polygon, Cardano, ...). As far as I know, there have been no successful attacks based on PoS's theoretical weaknesses, like its reliance on weak subjectivity. Some PoS networks like NXT have been in production for nearly a decade now. Isn't that fairly strong evidence that PoS works fine in practice?