CSV files are simply lists of tuples. You can do lists of lists in JSON. Your overhead will be two characters (opening/closing braces) per tuple (one if you skip the newline, which you can't skip in CSV), plus a dozen characters for the whole file. I don't think that's a lot of overhead, seeing that you get the benefits of a format that is actually standardized.

Example (CSV):

    name,age
    James,32
    Nina,10
    Helga,90

JSON:

    {"fields":["name","age"],
    "content":[
    ["James",32],
    ["Nina",10],
    ["Helga",90]
    ]}

Don't tell me it's harder to parse, cause you'd have to do the index→fieldname conversion in CSV too.