Even if somebody took the time to fake a ton of data to “prove” drop rates change, that dataset would still be an outlier since there have been hundreds, if not thousands, of users that have shown the drop rates to average 33%. Outliers are to be questioned as they fall so far outside the norm.
If I had 1000 quarters and flipped them all 1000 times, you’d expect 500 tails and 500 heads on each one. But if 2 of those 1000 came up 900 heads and 100 tails, you’d definitely have a look at the quarters to see if they’re tampered with or just plain fake.