For Intel's Tick-Tock model, the first move from one lithography size to the next generally has very little performance-per-clock difference in the instruction set. However, smaller CPUs allow for less power use, and will let the creator decide between shrinking die size or adding more cache/GPU compute nodes/etc to make up for the extra space they have. With overall-smaller die size reduction, they'll create more chips per wafer which should reduce the cost per chip. The reduced power draw will allow for more energy efficient CPUs and higher clock speeds within the same power envelope of the previous generation.
In short, smaller CPUs are generally cheaper and faster, while newer CPUs will gain incremental advancements in the instruction set and additional features, such as AES encryption acceleration. And in the world of highly integrated CPUs nearing system-on-a-chip, you'll have updated GPUs, memory controllers, PCI-express controllers, etc.
But again, with Intel's Tick-Tock method of first trying out a smaller process size on a known CPU design before upgrading the architecture to really knock our socks off, we've generally been seeing the first move to the new process that current software is within +/-5% of the speed than on the previous generation CPU when compared at the same clock-speed, while being more power efficient though.