Aah OK, so it has been suggested before.
I thought uniform randomness in occurrence of alphabets => higher entropy => harder to compress, and vice versa
(For instance if it's known that I would never use 0 in a binary string, I just need to specify the size of the string, rather than the content).