My blog entry on why I like HTTP URIs rather than URNs provoked two comments, both relating to the DNS basis of HTTP URIs.
There are 2 main scenarios where a domain name is transferred
1. Organization A owns domain a.com and creates a namespace at http://a.com/ns/foo. They put a RDDL document at that uri.
2. Organization B buys organization A. They decide that the foo ns is really about something else (p**n perhaps). They change the document.
How is this incorrect? The owner of the identifier has changed the meaning of the identifier. My guess is that the use of documents that use ns foo are going to plummet well before step #2 happens.
If you use an XML Registry for a urn instead, the exact same problem will happen. Org B has the right to change the "address" for the foo urn.
Domain transfer without buyout
There is also a scenario where domain names are transferred to Org B without Org B "buying" Org A, potentially through mistake on the registrar. How many times has this happened in real life? And how many times has the transfer of a domain name happened from an organization that would ever mint an xml identifier?
It seems to me that the answer to the question "How many times in the history of DNS has a domain authority been (by varying definitions) erroneously transferred where the original authority is a reputable organization that would or has minted a urn uri" is in the very low single digits.
And if that's true, it seems very dubious to create new schemes (like XRI) and use URNs based upon such a few erroneous cases.