While comparisons between the two versions are interesting I find the most interesting thing to be that the hardware used in the test is small and common enough (ignoring 15k SCSI drives and the SAN) that it should be possible to reproduce these tests at customer sites and get similar results.
So how can one use this? Well, whenever someone complains about BizTalk performance you can start by helping them set up a test some of these simple tests and compare throughput to the numbers in this document. If you find big differences you know to look at the server configuration as similar hardware etc. should give similar results. If the performance on the other hand is near those quoted in the document you can conclude that the server configuration is ok and can start looking at the design of the solution.
Even though the report is not as detailed as a full TPC-C report (3 MB example)(but who really wants that anyway?) is does include useful information about how to increase HTTP connections for BizTalk and also a sample on how to poll data from SQL Server reliably (this was once a favorite topic of mine, so I think I will have to follow up with a separate post on this).
Please note that the throughput quoted in the doc is sustainable throughput, read Wayne Clark's explanation of this.
EDIT 2006-08-28: Added link to whitepaper...