The external interface needs some tweeking for consistancy, but mostly I'm rewriting the internals. Currently it's rather crude and uses glBegin/End for all the objects, with no state sorting at all. I'm going to rewrite it to use vertex arrays and add batching based on state and texture use, but if it all goes to plan it should still look exactly the same and use the same interface.
First off, vertex arrays. Which always annoy the crap out of me because they're much more awkward to setup and use in Java than the C equivilent, but should make things much more efficient. Not sure about extending that to include VBO support yet, from some crude tests a while ago I found that for 2d stuff they didn't offer any speed gains over vertex arrays.