It's a matter of trust. A random instance can always lie and you can only determine "that was a malicious instance that was lying to me" in hindsight after it's broken that trust. Since a malicious instance-runner can spin up new instances almost as easily as creating new fake accounts you end up with a game of whack-a-mole where the malicious party can always get a few bad actions through before getting whacked. Whereas if user account creation was recorded on a blockchain you don't need to ever trust the instance in the first place. You can always know for sure that an account is X days old.
A malicious instance-runner could still spin up fresh instances and fake accounts ahead of time, but it forces them to do it X days in advance and now if they want to keep attacking they have a longer delay time on it. A community that's under attack could set the limit to 30 days, for example, and now the attacker is out of action for a full month until their next crop of fake instances is "ripe." As always with these sorts of decentralized systems there's tradeoffs and balances to be struck. The idea is to make things as hard for malicious users as possible without making it harder for the non-malicious ones in the process. Right now the cycle time for the whack-a-mole is "as fast as the attacker wants it to be" whereas with a trustworthy account age authentication layer the cycle time becomes "as slow as the target wants it to be."
That'd be nice. Personally, I think the tech is just about ready - Ethereum has solved its environmental issues with proof-of-stake, and has solved its transaction cost issues with rollup-based "layer 2" blockchains. At this point I think the main obstacle is the knee-jerk popular reaction to anything blockchain-related as being some kind of crypto scam. I'm actually quite pleasantly surprised that I haven't been downvoted through the floor for talking about this here so perhaps there's a light at the end of the tunnel.