I'm not entirely sure about the reliability of my logic here, but I did a little research and I believe what the National Review article is referring to is the 284,000 ballots already cast, in comparison to the 1.5 million votes cast in 2008, which results in approximately 18%. That's actually correct, as far as I can tell. However, there's a huge caveat here: more than 463,000 people have requested ballots, which is more like 30% of the 2008 total. The party breakdown of these requested ballots is 45% Democrat and 30% Republican. Assuming all those ballots are returned, that's a distinct advantage here.
Of course, I think that we should be skeptical of Marist/NBC as a true representation of the eventual outcome, just as we should be skeptical of all individual polls. What the National Review should be more worried about is the fact that 538 gives Obama a 73.5% chance of winning Iowa.