http://software.intel.com/en-us/articles/intel-nas-performance-toolkit/
Would that be of any help (32-bit only, though?)?
Are you just trying to prove that it really is this slow, or trying to find where the real problem is?
Are there any other network services that can be tested (such as basically doing a local DoS on a web interface, and checking performance/load during that)?
And, you're also 100% sure that all drives and arrays in it are in tip-top shape (I would hope a big vendor solution would annoy the Hell out of anyone near the server room if any weren't, but...)?