Not quite almost anything imo. My analysis could be wrong, but the binomial distribution should apply here.
http://people.hsc.edu/faculty-staff/...alPlotter.html
100 trials:
You should basically always finish between 35 and 65%. For simplicity I'm assuming no skill (p=0.5%), but the generator does allow p value specification.
Here's 1000 trials:
Now the outcome is almost always between 45 and 55%.