Not sure if there are any comparative studies about it, but aren't random numbers a lot more readable than random alphabets (with numbers)? I have read people raise this issue with IPv6. Choosing to use just numbers seems like a good idea to me.
I can assume that they fell out of favor because human memory favored the "3 digit-4 digit" chunking of the phone number rather than the "two letter-5 digit" chunking.