There was a startup featured on HN a few days ago (forget the name) that aims to do similar things. Essentially, they wanted to provide native linear algebra and associated data analysis tools for large data sets by providing the user with a Matlab/R like interface, and running appropriate Hadoop/MapReduce code in the backend. They are backed up by some of the biggest names in the machine learning community, including Tibshirani of EOSL fame.