The way the scope resolution test works, JSLitmus is getting normalized results by subtracting the time required to do this:

while (count--) {}

… from the time required to do this:

while (count--) {x++;}

… to get the time required to do “x++”. In non-SquirrelFish browsers, resolving “x” takes a lot longer than doing the “++”, so provides meaningful data for how long the resolution process takes. But in the webkit builds that use SquirrelFish, this code is all compiled down to bytecode, which eliminates the variable resolution overhead, so JSLitmus is just measuring the time needed for a native “++” operation. That operation is so ridiculously fast, however, that if there’s any additional overhead associated with the JavaScript interpreter, it will affect the test result accuracy.

CloudDream posted a comment on our zenbe.com blog showing that he’s getting “infinity” operations per second for Google Chrome, which I suspect is doing something similar. The “Infinity” result means that JSLitmus simply isn’t able to detect a significant difference between the calibration loop it uses (the top loop, above) and the loop that runs the test code.

Safari 3.1.2: http://i37.tinypic.com/nyi7u1.png
WebKit nightly: http://i38.tinypic.com/34hbcjn.png

That’s over 10x performance increase!