What this means is that a player can improve his expected return from -1% to -0.05% (!) using the above (truncated) martingale strategy.
This in turn means that the current 0.5% max winnings is at least 10 times the optimal Kelly bet against a martingaling player if I'm not mistaken. And that is dangerous.
I just started writing an "oh my God you're right and it's even worse than that!" post:
Not only that, but as well as reducing the effective house edge, the martingale player is also doubling the effective max bet. He's never betting more than 64 in my example, but is effectively betting 127 to win 1. So he's almost doubled the effective max bet from 64 to 127.
But... we don't have a max bet. We have a max profit. Kelly tells us not to risk more than 1% of the bankroll per bet. The "all-in 127 to win 1" bet profits by at most 1 BTC. But the equivalent 7-step martingale bet profits by 64 units on the last step. I think this factor of 64 difference means I need to scale down your statement, to:
This in turn means that the current 0.5% max winnings is at least 10/64 times the optimal Kelly bet against a martingaling player if I'm not mistaken. And that is safe.
I'm not convinced - and you did have me worried for a minute there, but I think the fact that the equivalent martingale sequence needs a much higher max profit than the single large bet it's replacing means that Mr. Kelly is still satisfied despite the reduction in effective house edge.