I cannot help but think that under a sufficiently precise set of assumptions (prior knowledge about the various one-armed-bandits, cost of losses, gain of winnings, expectations of being still in business in the future, etc), there is a provably optimal strategy.

Isn't there?