I searched on the archives and read the replies to the question asked by “sheel.at” on the subject. The final answer by Patrick - is that the end of the discussion? Is this still under review?
There were a couple of other threads on this on the ACME mailing list in the last few months. I think this one is the most recent.
The consensus seems to be to leave the spec as-is for now (to get it to RFC status soon), and possibly add a new endpoint that would allow disabling authorizations held by other account keys once the owner of a new account has demonstrated control over the relevant identifier (DNS name), or something similar, in a future spec.