If you are dealing with big datasets and need to find data within that set quick and easy, XML would probably be faster because it would parse it natively, and using native xpath-support it would probably find your data blazing fast.
Using JSON in those cases would require an eval on the data, and most json-parsers also validates a lot of the code for security reasons etc meaning that it wont be exclusively "native" code that processes it.
That will make a huge impact on the total processing time which would be a lot of overhead if you would only use a small sample in the data.
Both techniques has pros and cons, so it's more dependent on the actual case it will be used in.