I saw in Optech's most recent newsletter (#1507012) that there is now a proposal for Testnet5:

Testnet 4 included mitigations for an issue known as the block storm attack (#528804) which could render the whole network unusable. This led to a depletion of block subsidies, which made it hard to acquirecoins for testing. However, Testnet 4 still retained a modified version of the difficulty exception rule with the aim of allowing CPU users a limited path to acquire coins for testing, to mine non-standard transactions that other miners would not relay, and to keep the chain moving if a large source of hash power were to leave the network. Shortly after Testnet 4's introduction, the exception has been systematically and persistently exploited, which prevented the exception from achieving the intended goals. While block storms were prevented, the network suffers from constant re-orgs of small
numbers of blocks due to multiple difficulty-exception blocks competing for the tip. This led to discussion about changing Testnet 4 to mitigate this issue (see mailing list thread for analysis and discussion).

You may recall that Testnet3 turned into somewhat of a mess when shitcoiners started trying to sell testnet coins and it became difficult very developers to get their hands on them, so in an effort to demonstrate the point that testnet coins are supposed to be worthless, Lopp took it upon himself to absolutely destroy Testnet3 with some mega storms of blocks.

Out of these ashes Testnet4 was created, but it has struggled with similar block storm issues (see mailing list link above). So the new testnet proposal is:

In Testnet 5 there is no exception to the PoW rules. This appears to be the logical conclusion, since any such exception could be exploited by a motivated attacker. This ensures the network’s behavior matches mainnet as closely as possible. BIP 54 is already enforced on signet through [Bitcoin Inquisition][signet-bip54] as of this BIP's creation. However, signet does not allow miners to test that their software reliably follows the rules of BIP 54. Testnet 5 provides a testing environment for this.

This seems like a reasonable path. I suppose the downside is that a developer who wants to acquire coins via CPU mining may have a lot of trouble. Also testnet with no exception to PoW might end up looking a lot like Bitcoin Lite.

The response on the mailing list seemed to be mostly positive. There was this interesting response from AJ Towns:

I'd argue that, economically, a pre-mine (or initial allocation/auction, whatever you want to call it) is probably helpful for a testnet rather than harmful, in that it provides a potentially large pool of coins that can immediately be used for testing, and to suppress the scarcity/value of mined coins. Not a blocker, just my opinion. Not pre-mining is probably the most defensible position from a regulatory POV, however.

I think being demonstrably willing to regularly create new testnet versions is probably also a good way of avoiding testnet coins becoming particularly valuable/expensive; so even if the technical reasons for a new testnet weren't compelling, that might be a good reason to do so on its own. It's been about two years since testnet4 launched, which seems like an okay cadence, if it were to actually become a regular thing.

Rather than dropping immediately to min-difficulty, having the difficulty drop by 50% every 120 minutes or similar could be a reasonable approach if it turns out, in practice that difficulty gets pushed very high by a miner testing new hardware, then block creation stalls for a long period, preventing difficulty from adjusting downward. Seems like something to defer to testnet6, though. (Could perhaps also just grab the dynamic difficulty adjustment logic that BCH/etc settled on, if something along these lines is necessary)

Fun times in testnet land.