The only reason SatoshiDICE isn't paying out 99% is because they aren't being squeezed by a more aggressive competitor offering a 98.9% payout. That day will come. Because with an increase in play, both SatoshiDICE and some other competitor(s) can still operate profitably on a lower house advantage without cheating (which isn't really an option as it would be exposed them and they'ld lose all credibility in a matter of days.)
Not true actually. SatoshiDICE doesn't pay out 99% because the luck/volatility of doing so is highly likely to result in periodic ruin for the house. I've worked with a few brilliant statisticians to come to this conclusion. Even at the current 98.1% house edge, there is non-trivial risk to the house. If a competitor came out with a house edge at 99%, I would not match it, I'd just wait for the site to bankrupt itself.
This is something we discussed with dooglus but he failed to appreciate, There is a tipping point with the house edge and bet range offered where martingale systems will start to work often enough to put the house at serious risk of going bust. The house edge is not a guaranteed earn and martingaling has the effect of reducing the house edge significantly (when the simulation data is averaged out of 10'000'000 rounds).
Since most Bitcoin casinos don't disclose their buffer size - how much they keep on reserve - there's a very real chance that houses will collapse if they let their volatility get too high. Players should accordingly be aware of the risk and not put too much coin into a casino on a long-term basis.
We initially used simulations to calculate what bets we can afford to run, now we use a combination of our simulation data and automatic maximum bets.
It is a configurable setting for each of our games (which Ive been ordered not to touch), each game we can afford to payout 100 successive maximum wins on maximum bets (codemonkey adjusts this setting manually for specific games to allow higher bets). This in our opinion is the only safe way to operate, if your casino cannot withstand an epic lucky streak as described you could be looking @ critical failure.