To repeat, you either earn dust with losing streaks that are too long or end up burst pretty soon. I guess with 0.4% win chance and a losing streak of 2800, it would be equal to a win chance of 0.4x100=40% and a losing streak of 2800/100=28 rolls. But 28 losing rolls is not worth the risk as at any point you risk your whole amount while earning only dust when you win. Basically, no matter how you tweak the odds and amounts, it can always be reduced to these simple numbers (to make things easier to understand)
It can't be guessed, you have to experience it by play for real and study the system from it real outcome. This system you can't reduce any percentage to another percentage by ratio, the most important part is: some numbers tend to repeat itself more than other number for example, this come from my real statistic summarization: 9975 would appear much much more frequent than 9976 or 9974, 9999 would appear much more frequent than 10000 and 8888 would appear the least frequent, i see 10000 appear few times but not any matching count of 8888 in the collected statistic (~ 20millions result).
I hope you understand that that was just a coincidence. If you were conducting your study on physical dice you could probably come up with some useful data because indeed physical dice, due to their imperfection, can land on one side more frequently than on the other. But math is perfect, and since all the game in question is based on it, all outcomes have equal probability
Well, that in fact remains to be seen
I don't mean to say that the case described here has anything to do with what I speak of further, but it is in fact a real problem with "math being perfect" as in real life it is far from being perfect. In real life we are dealing with physical devices like random number generators which may or may not be entirely random (as with physical dice you refer to in your post). And I'm not even speaking of purely software ones which are called pseudo-random for a reason