I'm not sure that a block chain is required at all
What prevents double-transfer of names, then?
The Namecoin v2.0 rules.
You have to understand that what Bitcoin provides for a "Namecoin v2.0" system that works on top of the Bitcoin blockchain is a (proof of) data publication and ordering service, nothing more. It's the agreed upon rules that would take that data and determine what it actually means in terms of who owns what domains.
but if it is, it's probably best to use the Bitcoin block chain.
That is dangerous because of the developers'
active hostility towards attempts to store non-transaction data in the chain.
it should be allowed (with appropriate fees), and in practice it's probably difficult to prevent it.
Unfortunately it's very easy to prevent and
has already been discussed on bitcoin-dev. It can be implemented the same way P2SH was added: a miners-only upgrade that is forced on the minority once the majority adopts it.
You might wake up one day to find that the currency you'd been using to store key/value pairs is no longer accepting updates.
Gregory Maxwell's P2SH^2 system is very clever, but it still doesn't prevent publication of data in the Bitcoin blockchain, only the Bitcoin UTXO set. There's absolutely nothing we can do to prevent people from publishing data in the blockchain other than miners using blacklists, and even that measure can be circumvented if what you are doing is timestamping; I wrote up a good Namecoin v2.0 proposal on #bitcoin-wizards a while back that I called "zookeyv" that is genuinely impossible to block short of having all Bitcoin users sign their transactions with real-world identities that a majority of miners verify.
I agree with theymos on this: the next namecoin should be implemented on top of the bitcoin blockchain.
FWIW I wrote an improved version of the "blockchain data uploading" script that was put in the Bitcoin blockchain a few months ago to get a better understanding of the counter-measures possible; I'm increasingly thinking it'd be worthwhile to publish that code in the same spirit of publishing security exploits. Better we understand these issues now than find out the hard way later.