I don't want to be too quick to throw away my guess but, you have a good point that the 45 is too low to change it to a 64. Its also not crazy that they would still use hex values in Shenmue for a lot of its background math though. Games like Super Mario 64 still did RNG that way and that was only 3 years prior. I'll also add that Shenmue is a very ambitious game. Good minimalist memory use may have been a high priority in the studio. Its also just good programming hygiene to not use more memory or cycles in a variable than you need.
It's not crazy, but it's also unlikely.
You give Super Mario 64 as an example, but it's a launch title for a console that's a whole generation earlier. It's hardly a fair comparison.
And what you say is generally true, but given the stores are their own self-contained areas memory shouldn't have been an issue especially not for the single-use random number generator (that actually doesn't need to take any extra memory - define the list as a constant and have the compiler hardcode the values, since they don't change anyway).
Its tough pinning down the proper stat for Gold Dural too. Because of its rarity the percived probability can get out of control really fast. Take this example:
Pretend for a moment that the actual value to find Gold Dural is 1/1000 and we each set out to observe the statistic.
We each agree to use the same save file and play the lottery 5000 times.
I get a stroke of slightly poor luck and only get 4. My percived probability is 1/1250.
You get a stroke of slightly good luck and get 6. Your pecived probability is about 1/800
Even in this scenario that we agree is more optimistic than reality, with a high sample size and little difference in outcome we observed a nearly 500 change in denominator from each other. This is why I want to point at common binary values. Mostly I just want a well painted target because then, even if I'm wrong, I can still measure how far off mark I am.
4 wins out of 5000 gives a 95% chance the probability is between 1/3 million and 1/625.
6 wins out of 5000 gives a 95% chance the probability is between 1/4529 and 1/459.
The 1/1000 figure is well within both ranges, because it is reasonable to get 4 or 6 wins out of 5000 at a 1/1000.
Your suggestion of 1/64 for 2nd prize would be like giving a probability of less than 1/3 million for Dural Gold - it simply doesn't fit the data I have at all. It's not consistent.
Furthermore, saying "a nearly 500 change in denominator" is irrelevant - there's a lot more difference between 1/1 and 1/500 than there is between 1/1000000 and 1/1000500.
I've won 2nd prize 72 times from 3360 attempts. The chance of this happening with a 1/64 probability of winning is minuscule.
Similarly, the Dural Gold at a 1/1024 probability the chances of not having won any by now are minuscule enough that it can be ruled out.