I'm currently dealing with CSV a lot as well. I tend to use the format=map approach but not nearly as large as 22 MB CSV yet. I'm wondering if, or how much more efficient it is to deal with this type of data as arrays and map data structures versus XML. For most processing I can leave serializing to XML to the very end. And if too large I would probably also chunk it before storing the end result.
Intuitively I would think that dealing with CSV as maps/arrays should be much faster and less memory intensive. --Marc
