@sivan: It's actually fast enough (except shadowmapping ) but the code is veeery dirty and yeah it's quite difficult to use at the moment.

Rewriting the pipeline would solve all of these problems and I'm sure it will be much more efficient aswell.