I have the top five as:

; (12 times) I (11 times) * (10 times) - (10 times) 2 (10 times)

and for the quick-brown-fox sample he posted in the comments, I have:

! (14 times) # (13 times) C (11 times) , (11 times) U (10 times)

So I'm not convinced this is a fruitful path of inquiry.

I doubt this information will be helpful since for every character there are ~9 characters after encryption.

you are on the right track.

Other observation: the encoded string's characters seem to range from ASCII 32 to 90.

That range is 58 characters which (perhaps suspiciously) is size of alphabet + 2^5. (Although a whitespace character is also encoded, making that suspicion a little less "round".)

From comparing the multiple encodings of Quick Brown Fox, the value distribution across that range is fairly even and every value is used at least once.