Sadly, everyone. It's less a matter of "slow GPU" as it is of publishing GPU bound benchmarks in CPU evaluations. We already know that half the processors released now will get the same fps at high resolution with the same processor, so publishing that info is just pointless, and serves to obfuscate the actual performance differences between procs. In my opinion, it's of the same value as putting in network benchmarks or disk transfer benchmarks with a 7.2k rpm disk in a processor review. You're benchmarking the GPU, not the CPU when you do that.
edit: Ok, my beef is really with people who then use those GPU bound benchmarks to make claims about the relative performance of the processors less so that the reviewers themselves 😉