Defining an effective threshold for "as high as it needs to be" is the problem. And this is well known. PoW was abandoned as a spam deterrent in emails precisely because no threshold could simultaneously block spammers (especially those with access to server farms) and remain practical for legitimate users (particularly considering smartphone and low-powered devices). And for emails it was actually easier (deterrence was against spamming millions of mails, not just sending few hundred events to saturate a relay). For me it is really surprising to see PoW being tried again in Nostr: history shows it is not going to work.