Depends what you are saving, but I'd have a hard time justifying it for most information. Maybe for direct-coded images, digitized analog input, or other pure data. If you get to the point where binary compression algorithms give reasonable compression of the stored data, then external storage would be far simpler to implement.
Not being able to directly inspect the values for even the simplest debugging. There are so many string, integer/signed integer, double, coordinate, handles, and other storage types that using binary doesn't seem worth the bother.