It's finally there for all of you map/reduce lovers (if you've heard of Disco or Dumbo or Mahout you know who you are), the Million Song Dataset has been added to the AWS public datasets!
It contains the original "Echo Nest core" of the MSD, but the additional datasets are small enough to be redownloaded if needed. Yes, getting started will probably require some code to transform all the HDF5 data into json or xml files easy to map/reduce, but any lab with small computational resources can now play with the million songs through AWS.
Thanks to Infochimps for making this happen!
--TBM
P.S. the official intro paper to the Million Song Dataset is now available [pdf].
P.P.S. Happy 6 months MSD!