I don't claim to understand how their particular hash generations work but it seems to me like you're right in that you can't use forum posts as a statistically valid way of counting winners. It is only potentially indicative of something and doesn't constitute as proof of anything.
Also, 20x50 is 1000 posts and is not statically relevant as a number because we have a very low % of winning numbers (under 1%), we need a bigger number to get a realistic value besides variance.
Does this website haven't a public list of rolls?