The EnsEMBL converter seems to be done now. Now to let it chug through the whole database (slow...), and see if there are any special cases I've managed to miss.
I'm now writing a Java bytecode generation library so that we can use dynamically generated classes in BioJava. To my surprise, there didn't seem to be a complete one available under the LGPL, so it's more coding for Thomas. Hopefully have some kind of release ready early next week.