It has been a long time! My memory ..............
Most 8085 boards only had 1,2 or three I/O addresses in use. (serial, parallel) I cannot think of a case where any one would need all 256 addresses. It takes MORE parts to decode 64000 addresses. (money) If you made memory mapped (16 bit I/O addressing) then there will be a hole in the memory addressing.
ALSO
Instructions: The CPU is 8 bit. So thinking 'do something at I/O at 5' is simple. If you need to do a 16 bit address for every I/O the CPU has a hard time: Get the high 1/2 of a address and put it in a register pair, get the lower 1/2 of a address and put it in a register pair, now do a I/O using the register pair as a pointer.
The CPU is 8 bit. If you want fast math you do it in 8 bits. Math in 16 bit is (much + much, test for carry, add the carry) slower.
I know my examples are bad/funny but don't do anything in 16 bits if you can find a 8 bit way. (hardware and software)