Same here. Some problems in ML are embarrasingly parallel like cross validation and some ensemble methods. I would love to see better support for Spark in scikit-learn and better python deployment to cluster nodes also.