We do have Otsu, and I think it would be within the scope of scikit-image.
I think I'd find use for this, personally, and depending on the implementation you may be able to reuse code from simple binary Otsu internally.
Seems reasonable - PR would be welcome!
Josh