First, they don't say wether IE7 is using it's compatibility mode or not, which could seriously change the results, probably bringing it right near IE8 and IE9.
Second: javascript benchmarks are meaningless. If they are going to test browser speed, they should test browser speed, not the speed of their javascript engine. Additionally, considering that all versions of IE use the jscript.dll that is part of the OS for javascript interpretation I find it hard to believe that the numbers would be so skewed.
Browser speed should be tested by disabling client-side script entirely and seeing how quickly it loads a page containing standard HTML.
Also, the sunspider test states that it is a "real-world" test, and yet, it says it doesn't test the DOM.
The main thing javascript is used for in the "real world" is for interacting WITH the DOM. therefore if one browsers DOM implementation is faster then another, this will influence the results of <real-world> browsing. it additionally states that it doesn't perform "micro" testing. and yet at the same time I witness tests that measure the speed of the bitwise operators. something I would call "micro-testing".
using a javascript speed test as a measure of a browser's speed is like measuring the speed of a game's sound subsystem to determine framerates.
