That's an easy one for me - text files.  I can read a 1 GB text file into staging, normalize a few things and write it to a permanent structure faster than the query to parse the XML would take to run.  The files aren't as huge, it processes faster and it's relatively simple to do.  Compared with XML, well...nevermind, there's no comparison.  The problem with text files is that they aren't shiny, new, cool or hip.  What some people call boring, I call efficient.

To the point of the excellent article, it can be difficult to renormalize data once it's denormalized into a structure for transfer.  Granted, it can usually be done, especially if you're working with the group that's sending it to you, but it can be cumbersome.  But when you stop and consider some of the stuff we do in T-SQL and how fast it runs, it is pretty amazing.  For a historical perspective, most of us routinely carry around a flash drive in our pocket with more data than it took to put the first man on the moon.