This is interesting, it is possible after all that pseudo-random numbers (for seeds) are not quite random. Why didn't you share full test results? It would be nice for someone to repeat it (though it's quite time consuming: 500 hours).
(btw, I've never rolled more than 9993 on any of my multiplier or regular rolls on neither faucet, probability of which should be over 75%...)
But you said you made only 500 rolls? Or did you mean some bigger test? You need more rolls to get 3rd and 4th prizes (at least 1250-5000).
[0 - 9885]: ~ 1/20000 + 9885/10000 = 0.98855
[9886 - 9985] ~ (9985-9886+1)/10000 = 0.01 (~ once per 100 rolls)
[9986 - 9993] ~ (9993-9986+1)/10000 = 0.0008 (~ once per 1250 rolls)
[9994 - 9997] ~ (9997-9994+1)/10000 = 0.0004 (~ once per 2500 rolls)
[9998 - 9999] ~ (9999-9998+1)/10000 = 0.0002 (~ once per 5000 rolls)
[10000] ~ 1/20000 = 0.00005 (~ once per 20 000 rolls)
sanity check: 0.98855+0.01+0.0008+0.0004+0.0002+0.00005 = 1