Freaky, no?
The success rate was around 63% during the time I was collecting results, very high compared to the uniform distribution rate of 17%.
Theories on why it works are cheerfully accepted. Here's mine. Thanks.