The TPC-H has some tests in it that rule out many MPP systems (ie transactions and true ACID compliance). In fact I'm not sure that the TPC-H is a relevant test of these systems while these transactional tests are retained in the benchmark. These systems do not try to be transactional systems and do not require these slow sequential features.
If you want to see good TPC-H figures, apparently Vectorwise has just blitzed the TPC-H with a very significant result.
We are using Vectorwise on our MPP nodes and are finding staggering performance improvements. But like GP (although Luke would probably never admit this) we cant run the TPC-H legally either yet.