They can't select 8-bit int. Must be >= 16 bits. Trying to match the register size is only an option for bigger processors, i.e. a 32-bit processor should preferably use a 32-bit int and not settle for the 16-bit minimum even if the instruction set would allow same-speed operations on both sizes.

And selecting a too large int size would kill the manufacturer in any benchmarks.

It isn't hard to realize why Keil decided to support deactivation of integer promotion. The 8051 design spent a lot of energy on efficient bit operations, making it one of the worst ever 8-bit processors for handling 16-bit and larger integers.