Sorry, Mr. Mandelbrot, but I am apparently missing the point: what does commercial realease have to do with tests?
Now that's a pretty uninformed question. Suppose MS discovers a security flaw before final release (hard to imagine, I know!). They might have to rewrite the microkernel, core APIs, or lots of other elements before they release the product. This could easily result in severe performance degradations.
You won't find a single commercial benchmarking outfit that would release a benchmark on prerelease hardware or software. Besides, what's the point in making a comparison against something you can't buy? And you missed my sarcasm -- Microsoft compared an unreleased product against one which was at least a generation old and substantially slower.
By the way, if Microsoft wanted to make the comparison fair, they should release an Java ODBC driver for SQL Server. The comparison was apples versus oranges in that it was SQL Server vs. Oracle. On the other hand, if they really wanted performance, they should release .Net for BSB Unix and watch it really make hay!