I think everything has its time and place, even synthetic benchmarks. The only thing they bring more than the real-world (in some cases not all) is consistency. For real-world test, more often than not there is a degree of human error. Most reviewers are stuck with programs like FRAPs, [a VERY handy tool] which leave it to the user to specify intervals to benchmark. The bad thing is, with DX9, AA scores are all screwed up even when running DX8 games. This is another reason why reviewers are hesitant to travel to the next generation of technology. It is screwed up!
