You want your "smart" lamp to respond quickly to reduce power consumption and air time. Just a short sprint once is a while instead of a slower 8 bit cpu crunch

You want your "smart" lamp to respond quickly to reduce power consumption and air time. Just a short sprint once is a while instead of a slower 8 bit cpu crunching on algorithms designed around 32 bit operations. The wider ALU and registers are cheap. The fetch and decode logic is almost identical in area and uses less power because you can implement your application executing fewer instructions. Have a look at the code size produced by common C compilers for something like AES128 targeting a Cortex M0+ (or M33 in this case) and a ATmega. You'll see lots of short helper functions that depending on compiler settings are either called very often, inlined or even inlined and unrolled.