Re: 10nm
"At the end of the day what matters most of the time is the speed from the CPU to ram and bus"
It depends on the workload - if your workload is predominantly IO then maybe, but I would expect larger caches, wider memory buses or faster IO architectures to be the focus or performance increases (all are available from other vendors) if required for what is a general purpose CPU.
For compute based workloads, your performance depends on clock speed, IPC and how much time you can keep the CPU busy - caches and longer instruction pipelines will largely hide the actual speed of memory or IO from the processor. Again there will be exceptions that expose worst case behavior, but that's where software optimization plays a part.
For two generations of the same architecture from the same manufacturer, clock speed and IPC will tell a significant part of the story.
Given the process node advantage of the newer generation chip, you would expect either a significant jump in IPC to justify the clock speed drop or more cache/cores to boost performance with all those extra transistors.