Using a geometric mean would give the result that a browser that took 1 second to do task A and 1 second to do task B would score the same as a browser that took .25 seconds to do task A and 4 seconds to do task B. It's not obvious to me that this is the case; I tend to think it isn't.
(Test bug with string indexing acknowledged, but it only affects one of the sub-tests afaik. It will be fixed in a future version of the benchmark. There's a reason I posted it for public review before publishing any competitive numbers myself.)