I forgot, k-NN could easily be parallelized for very large data sets