Why not try a Monte-Carlo-Simulation? Just create random sequences of given length and amino acid composition and calculate similarities. Put the results into classes and plot frequency against value.
