Most of the literature I've read about GAs suggests using a crossover value of around 0.7, so you take the first 70% of one chromosome's genes, and the last 30% of the other to produce one new chromosome.
If you are picking the parent chromosomes by taking the top two (ranked by fitness), then I can see the logic here, as you are giving more weight to the genes of the higher-rated chromosome. However, if you're using a stochastic method (such as a roulette wheel) to pick the parents, then what's the point of using anything other than 0.5 as the crossover value? Given that you have picked chromosomes A and B as the parents, you're just as likely to pick A first and B second as B first and A second aren't you?
I've only actually written one GA so far (still way down at the bottom of the learning curve, but moving up fast thanks to some great help here!), but experiments on that show that 0.5 gives a faster convergence to the solution than any other value.
Or am I missing something?