Quote Originally Posted by MK27
If the distribution of characters in the English language means that using the mod 2 algorithm will on average yield 47% zero and 53% one, that means that by changing the algorithm, you could achieve exactly 50%. Just pick the characters according to the average distribution.
Not necessarily, e.g., it may be the case that there is no configuration that achieves exactly 50% on average. But yes, this is why I decided to describe your proposed RNG in terms of an unspecified hash function instead of ASCII value modulo 2: I felt that in a practical implementation it may be necessary to use a different hash function.

I do not think that this is necessarily enough though. My impression is that statistical randomness involves more than just a uniform distribution (after all, a uniform distribution can be converted to other random distributions, and vice versa).

Quote Originally Posted by MK27
Sure -- so in reality this would mean what, you start guessing all the keys starting with the 51% more likely bit? Horseshoes and handgrenades! What is that going to help?
I am not sure: my cryptanalysis knowledge is really bad. But remember, we're talking about the information theoretic perfect secrecy of a one time pad. This is uh, different from the reality of cryptography and cryptanalysis in practice.