CPU performance cannot be expressed as a single benchmark score. Like Dhrystone, CoreMark scores mostly depend on frequency. You'll get a far more realistic idea of performance from larger benchmarks like Geekbench or SPEC2017. Larger benchmarks require good branch predictors, automatic prefetchers, low latency caches, streaming optimizations, high memory bandwidth etc.
This reminds me of the good old StrongARM. It was very small, simple and extremely fast as long as all your code fitted in L1. So it had amazing Dhrystone scores (or CoreMark had it existed back then), but it wasn't very fast when you ran complex applications on it.