It turns out that investor amounts are not Pareto distributed (the continuous analog of Zipf's law), but rather a mixture of two log normally distributed random variables. I've posted the details here so as not to junk up this thread. I'll just leave one chart to give you an idea.