I don't think that's the case at all. Consider Levine's point about self-trading. Even algorithms developed within the same company have such different views th

I don't think that's the case at all. Consider Levine's point about self-trading. Even algorithms developed within the same company have such different views that one believes the price will hold steady or revert and expresses this view by sitting on the offer while another is so confident that it will move that it's willing to lift that offer. Just looking at ultra-low latency algos in liquid products like treasuries you have a lot of diversity: market-makers, yield curve spreaders, cash/future/ETF arbitrageurs, statistical signal trading, along with combinations of all of those. All of these players need to be fast but have vastly different risk profiles and trading style. If you made trading more expensive or slower, traders doing useful things like making quotes or keeping the cash and futures markets in line would be driven out, making the markets less efficient. I'm less sure about the value that high-speed momentum strategies provide. If they're right, they move prices to be more efficient more quickly, but they can also negatively impact liquidity providers and make prices inefficient if they're wrong. I figure the market will sort that out though. Nobody can run a money-losing algo indefinitely.