The question is why didn't the bot work as expected, ideally the experiment should have resulted in all the BTC being burned into the fee (or at least a lot more TX than 3 or 4).

It seems the mempool rejects some TX even if the fee is substantially higher. There's something alberto does not tell us, or he was just very lucky. Or maybe the mempool push TX api was the secret sauce that worked.

Some ideas:
- spam the mempool with multiple TX with different output addresses and fees
- broadcast the TX to multiple destinations

And I don't think miners are idiots. Why would they help some random guy for 10 or 20 or 50%, when they can definitely
play the game on their own?

It took TWO seconds to crack the private key. Another fraction of a second to create the new raw TX with higher fee. But the TX pushing failed because of some mempool rules, that was the weird part. Anyway, it's not a matter of minutes here. So any miner can run a bot, do the cracking themselves, and include it in their own mined block no matter what happened in the mempool.

With zero fees. The miners may also reject TXs that try to increase the fee from other people.