I don't think it would alter game balance, as you still have 20 outcomes with same probabilities, just as with default RNG. It is not weighted. Using only statistical tools (i.e. computing mean, variance, etc.) there should be no detectable difference between this and default RNG over some longer period (like 1000 samples). The difference will be visible over short period and that's a good thing because it eliminates the "bad luck" impression.
