I could not get rid of the url encoding and decoding without hampering some users ability to pass non-ascii characters back to JavaScript. However, I was able to reduce the amount of data being passed from Java to JavaScript by 40% by decoding common characters that occur in JSON and XML. These characters will survive the round trip just fine and don't need to be encoded.
This is the best solution I could come up with. You won't be able to read files as large as you could in 0.9.4 but it will get close and it will support non-ascii characters.
Contact search was not working for unicode letters. The CallbackServer was changed so that it returned url encode strings. On the JavaScript side the PhoneGap callback handler decodes the returned string.