I believe for a true random sample you would have to start with equal amounts of each headstamp. ie: say you have 300 cases, 100 would have to be win, 100 rem, and 100 PMC. Then you could draw 30 rounds for FSPM, replenish equal amounts and draw the next 30 for WSPM and so on...
But then it has been a while since that statistics course...when will I forget this crap? :-)