I've done sort of what you suggest professionally on projects where design decisions are driven largely by per-unit cost. If the SoCs that do everything you need aren't cheap/available, go with the best-fit main microprocessor to handle the heavy lifting, and one or two cheap micros to fill in the missing pieces. It's also a good approach for when a part of your solution needs good strict real-time software.

However for low-volume and/or personal projects, dev cost/time often trumps hardware cost and heterogeneous systems have a whole host of secondary challenges. Specific to your recommendation, it's a more complicated power architecture, more components to enclose, more tooling to worry about (software and hardware), and I have to worry about how to synchronize and communicate between the SBC and the micro/arduino.

It's worth an extra $100+ to be able to focus my limited free time on solving the problem I want to solve rather than on "shaving yaks."

Ah got it. Thanks. I didn't think power would play a significant role, but if it does then I can see why.

It all depends. Are you running off batteries, or wall power? Does your system have idle time that it can take advantage of for power savings? Were the boards in question designed with features to allow for low-power sleep, etc? Are you working with analog? Do your boards have low noise supplies, or do they expect a low noise input source? Are you doing any high current or high voltage switching? Is power-coupled noise an issue? Are we doing any switching of mains power? Is safety an issue? Should we use isolated supplies? If so, how much isolation do we need?

Software folks, myself included, tend to drastically underestimate the complexity of power design...

There is a reason an entire branch of electrical engineering is devoted to it. A branch I am considering going back to.