Consider the flash loan itself

 All the following contracts are on the Eth chain, and paid for with miner fees on the chain. The flash loan says you borrow the money and it reverts back entirely to the lender, always, within some timeout. Inside the contract are trades with buyers and sellers, also on chain. 

The borrower is thus the middle man using temporary risk capital to match buyer and seller and this is a verifiable profit, verified by the miners on Eth block chain! The profit sufficient to cover miners fees and return the fund. If the condition is not met, none of the intermediate trades are verified.

This is a pit boss function. The entire operation is based on one simple detail, the exchanges are all OTC on the Eth block chain, there is no pit boss risk function, the contract, in total, is performing the risk arbitrage. This, it turns out, is all about correcting bad Oracles that need to report market congestion, but do not. If you buy or sell using Oracle price information with no run time pit boss, then you suffer arbitrage, an unexpected market loss.  

The contracts seem to be working correctly, it takes time for the market to dump the current generation of Oracles for the smarter version.  Total arbitrage losses are still below 10 million, and that seems like normal. It is the same cost as running stock through Robinhood while the company sells you advanced information for arbitrage.  It is shit that happens when traders do not understand congestion, it has nothing to do with the Eth block chain except there are a bunch of traders with asymmetric information. 

Why don't the Eth miners do these trades themselves? A lot of them do. But the Eth block chain backs up and miners fees rise, like and interest charge, congestion pricing eventually holds. I would say that under the circumstances (that is, asymmetric Oracle information), the trade are quite a normal correction of bad Oracle code. I say let it continue, it is information about the problem and the trade causes the market to adapt.

Taproot, the contract coding byte code on Btc will interface properly with  Eth. The two ledgers can be made conditional on each other. Thus we get true, verifiable ledger swaps within contract. But this exposes the fails to deliver for pricing, a good thing. When ding this completely under one ledger, the fails to deliver are cancelled before being noticed. But, like the rest, we need the fails to deliver queue, it is a measure of congestion.