Further speed improvements are very difficult, indeed. I have been experimenting with a different algorithm, but that is (currently) still slower than the current one. Though there is still more room for optimization there.

I plan to release an update on the current algorithm that should bring a small improvement, probably by the end of the week.