Sure, I know and like DNSBLs including Spamhaus's, but this is a distinct application from XBL. Specifically, removal needs to be rapid in order for it to be useful for rejecting customer Web traffic. That's an engineering requirement that email anti-spam systems don't have, since SMTP is designed to retry for days if necessary to get a message through. Moreover, hosts that send any legitimate email are very few compared to hosts that send Web requests; and even though email admins are frequently dense, unresponsive, or victim-blaming, they're still a level above typical users in knowing what the fuck is going on with their computer.
One approach would be to have each DDoS victim continually (e.g. every hour) assert which addresses were attacking it, and only list those addresses which are currently attacking. This way, as soon as a host stops attacking, it will drop off the list. This has weaknesses — for instance, an attacker can use your host all night while you're not using it, without you noticing — but it's still an improvement over what we have today. And it still depends on each subscribing site having a good enough backchannel to the listing service to stay open during the DDoS. Back in the day we'd do it with a dedicated modem line — the bandwidth requirements are really quite minimal — but nobody knows what that is any more.