It's a LOT of effort. And said effort is required not just at the registry level, but also at the dozens of registrars that support EAP.

This isn't a theoretical concern. We actually did use standard auctions for our first three TLD launches. They were operational nightmares. The Dutch auctions we're using now are much more light-weight, have greater participation, and are now the preferred way of doing launches across the entire industry. We're not even the ones who started using Dutch auctions for TLDs.

You'll just have to trust the word of someone who's been doing this stuff for years that the alternatives really are worse.

The thing is that DNS infrastructure (which registrars basically rely directly on) is very OLAP oriented. bind(8) and DNS daemons like it are DBMSes in some sense, with replication and serving highly-concurrent reads being their primary focuses, and inserts/updates/deletes taking about fifth place compared to those needs. And registries (and sometimes registrars as well) build their entire infrastructure to use the DNS-daemon “store” as the canonical store, rather than having it be a secondary system synchronized into from an online OLTP DBMS. So this writes-are-expensive paradigm creeps into the entire DNS infrastructure, including things seemingly far away from the core, like the registrars’ control panels.

A good comparison is blockchains, which are also highly-replicated, cheap-reads expensive-writes DBMSes. CryptoKitties chose Dutch auctions to sell their kitties for a similar reason to that of the .dev registry: since putting state into their “DBMS” is so expensive, they wanted as stateless a system as possible. (In their system, even the current price is just statelessly computed by comparing the client’s time-at-bid to the chain’s recorded start-of-auction time. The only writes needed are to start the auction, to place the winning bid, or to cancel the auction.)