Configurable variants distribution

Currently the set and variant occurence is equally distributed.

We want to be able to tune the quotient and make e.g. Variant B happen 5 times less than Variant A and still have "right" statistics.