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
The client seeds are actually generated using javascript entirely on the client side. This is the part of the Javascript code that generates the client seed (can be found in the JS file