> Or overestimated the capabilities of the developers?
That could very well be, too.
Although I most often experienced the former issue, project managers going as cheap as possible. And not only once have I seen the grudging "upgrade" to a bigger, more expensive MCU variant, more often for code space issues than performance issues.
But I think this is a critical point some developers and project managers did not understand. Cortex M is general-purpose application architecture, not optimized for high-throughput (like DSPs). High data rates and interrupt frequencies (like for cycle-by-cycle control) can quickly stuff the core, and eat up 80...100% processing time. And no matter if you throw more expensive hardware (M7, dual core) at it or more specialized / optimized hardware (DSPs), it will not come cheap.
