Working with a mid-range ARM, I noticed how many configuration bits and bytes they have (I am spending quite a lot of time just setting them or making sure they are set properly by default).

Every such bit means a flip-flop, consuming at least as much transistors -> silicon area than a SRAM bit (and probably quite a bit more of the silicon area, the SRAM being packed with maximum efficiency).

Thus, such a mid-range ARM has only in those (in the vast majority of applications never used) configuration bits more transistors than the whole SRAM in a contemporary mid-range 8-bitter, including most '51 derivatives...

JW