A question of statistics

[Quantster]

Hi everyone,

I've been working at a betting model lately and I've came up with several variations on it.
Can you help me which version would be the best? (see picture attached)

The model is applied over 16 events, and the pictures include the whole distribution over all 2^16 cases.
This emulates very well a coin flipping process.

Now the question is which strategy is better statistically. What's the most important? Given the same average, which is better to pick ? Very low average loss compared to average win at about 80% win rate, or high average loss compared to win at about 96% win rate? Also the quartiles are not to be ignored.

=Even later edit=

Mistake in the MM function. Distributions have to be made again. However betting logic still stands.

The system is not applicable straight to roulette, because of its limitations:
1. It doesn't support roulette 0 (it's not included in distribution, and it's probably fatal)
2. It needs a high max bet/min bet ratio for outside bets.
3. The amount of calculations is too much to be done manually in a casino with pen and paper. It would require a smartphone/tablet application.

[Quantster]

-