If that was easy to do, it'd be done. So without someone going through the trouble of changing internals, and, as importantly, without incurring measurable overhead, this probably will not be done with default implementation(s) (there are separate backends for byte[] and char[] based inputs).
Parser does not keep track of content to decode beyond current token, and with streaming amount of content in underlying buffers varies.
But while it is difficult to do this in general, for specific use case you can either implement buffering input source, or other types of buffering. When you know the exact use case and input sources, generally tricky problems become easier to solve for specific usage.
-+ Tatu +-