API versus downloadable dataset

There's an interesting discussion on Nat Friedman's blog about API versus downloadable data. Lots of new data sources are popping up online that wrap up their databases in a nice friendly fashion, which is great, but sometimes it would be very useful to cut out the middleman and grab the databases directly. Mark Colburn brings up some reasons to prefer an API, including clarity of provenance, reliability of data, and maintaining control over updates.