Most political polls, which are designed to measure the opinions of hundreds of millions, only use a few thousand people as a sample size, so this study probably has a very large predictive value.

You do have a point that the gathering of the data may be flawed based on who was more likely to answer the poll, but there's nothing wrong with the sample size.

"The sample size is invalid,"

This former statistics TA would stamp an F on this statement, with good cause. 2500 people is a very good sample size for a wide variety of topics of study. The only time you need significantly larger numbers is when there is a very tiny difference between subjects, or if you are studying something (usually a vaccine for a disease) that has a very small chance of occurring. I would say that virtually all medical research that has ever been done has used way less than .025% of the human population, and sometimes the numbers for human trials of medications and treatments can be strikingly small. However, they are still valid and reliable sample sizes (as evidenced by the gazillions of drugs we now have in Western medicine).

I like your point on standard deviation however. I would have liked to have seen a distribution chart (bell curve style) with the y-axis being the number of players and the x-axis being amount of gold. That's the one thing missing from that report IMO.