Thanks for your bug report. This has been an ongoing topic since $rand() was first introduced :-) I have changed the random number algorithm many times over the years. Unfortunately, every time I change it, someone eventually performs an analysis and reports that it is biased. That you've found the biases are minor is great. Usually, someone finds something drastically wrong with the RNG.
Currently, the RNG code in mIRC uses a Marsaglia variant. Also tested but commented out are Mersenne Twister and Wells512. I would have to look through older versions to list all the RNG variants I have tried. On my to-do list are PCG
variants. However, the odds are that all RNGs contain issues of some kind, which is why it is such a thriving field of research - the perfect PRNG has not been found yet.