Kevin Jones has published the result of a simple benchmark performed on seven XSLT processors, including three different tests to simulate different types of transformation.

Despite real life testimonies like the one of Mike Moore from Microsoft.com, XSLT is generally perceived as a slow technology--as recently summarized in a rant against XSL by Luca Passani:

Performance issues: need I say more?

In this context, Kevin Jones announced on the xsl-list a simple benchmark performed on seven XSLT processors (by order of decreasing performance: Napa, MSXML, Sablotron, XT, Xalan C, Saxon, Xalan J).

In the interests of fair play, Jones notes that Napa is a not-yet-released processor he has been writing:

Napa is an (un-released) engine that I have been developing. While I have tried to avoid using this benchmark during development its high score makes me suspicious. Until there is other (independent!) evidence available this score should probably be discounted.

This benchmark may be qualified as being "out of the box," as it doesn't involve any optimization. It was performed on a Pentium III 550Mhz with 196Mb running Windows 2000, and the Java VM used is IBM 1.1.8 (for the Java based processors).

Some suggestions to improve the benchmark have been posted on the xsl-list:

• Sebastian Rahtz suggested including Oracle Java XSLT in the tests, use different versions of JVMs to sjow the dependency of the relative performances against JVMs, and to use a more recent release of Saxon.
• Andrew Kimball from Microsoft suggested running the benchmark twice (first time "out of the box," and second time with optimizations) and offered to provide code snippets to show how compiled stylesheets can be used with MSXML.