All the benchmarks I have ever seen are like EPA gas mileage figures; artificial numbers the are optimistic at best. But numbers because of how they are derived have some comparative value at least in a gross sense. EPA mileage figures do not actually account varying load, weather conditions, etc. that affect the real world numbers. Benchmarks have the same problems, what is the proper balance between CPU load, memory load, drive reads, etc. to mimic the mythical average user? The 'crime' is not that benchmarks are skewed but that they researches are misusing them to try predict behavior in the real world when they are only good tell general trends.