Its very simple: don't use flawed benchmarks. For instance, consider the UI refresh benchmark. For Mavericks, it reports over 800 refreshes/sec while for El Capitan, its just 80. Surely it means that Mavericks is much faster, no? The crucial fact, which XBench blatantly disregards here, is that the display can only display 60 updates per second. This means that modifying the UI hundreds of times per second is completely and utterly pointless. In fact, an application that attempts something like that should be considered buggy. This is why OS X introduced an optimisation on 10.10, which removes all the unnecessary (invisible) updates. For the user, nothing is changed, but a lot of unnecessary work is avoided — the buggy application uses less CPU, your computer runs cooler and the user is happy. The same probably applies to the Quartz drawing tests as well.
XBench might have been useful some years ago. However, at this point it should be retired to the garbage dump of history, where it rightfully belongs