These puzzles are called pangrams. Douglas Hofstadter wrote about them in his Metamagical Themas column in Scientific American in the early 80s.

Someone came up with an algorithm to solve pangrams, and it involves getting the values right letter by letter. First you plug in any old value for evey letter ("This sentence contains one 'a', one 'b'... and one 'z'). The you adjust it so that the 'a' count is correct. Then you adjust it so that the 'a' and 'b' are correct, and so on, until you reach 'z'.

Maybe you could adjust your measure of fitness to take this into account. Other than that I don't have any remarks on your code, which looks pretty fine to me. The only thing I'd suggest is that .3 for mutation is very high.