Just consider A8->A9->A15->A57. Geekbench single-threaded result increases 10x from 600MHz A8 to 2.1GHz A57. That's a 42% IPC gain per generation plus a 52% clock frequency gain. Are you denying that?

Most benchmarks exhibit more ILP than are available in commonly used programs. They are not representative of lambda user's workload. These results are impressive (and that's why industrials use such benchmarks to praise the quality of their products) due to the very nature of the workloads: a lot of parallelism and predictable branches.