Some improvement to the algorithm is still possible. scikit-learn's implementation of K-Means contains some optimizations that I think should be applicable for K-Prototypes as well.
Also @KristofMorva mentioned the paper below, which may or may not do something similar to scikit-learn:
http://www.mdpi.com/2073-8994/9/4/58/pdf
Some improvement to the algorithm is still possible. scikit-learn's implementation of K-Means contains some optimizations that I think should be applicable for K-Prototypes as well.
Also @KristofMorva mentioned the paper below, which may or may not do something similar to scikit-learn:
http://www.mdpi.com/2073-8994/9/4/58/pdf