My Kaggle partner was trying out Incanter and gave up due to performance reasons. Apparently, when Incanter loads a data set, it loads every cell as a hash as opposed to loading rows or columns as vectors. Correct me if I'm wrong - he would love to use it.