XBench is a fairly useless means of comparison, even when looking at specific benchmarks. It's probably less useful than MacBench was, whose code was small enough to fit on the backside cache of 604e and G3 computers with a full 1 MB of cache, reflected when the cache was OC'd- something which almost never affected real world performance.
They results are wildly variable, and even if they weren't, the tests aren't really related to real-world performance.