How can this be true consistent with HFT being profitable?
> The faster the trading, the lower the spread, the smaller the leap sellers and buyers have to take.
That's the theory. The issue is that when you get into sub-millisecond trades, you're so far into diminishing returns that the profits the HFTers are taking exceed the benefit of any theoretical increase in liquidity.
Also, the paper link in that post is truncated, but I found it on Google and (at least from the abstract) it doesn't seem to support what you're saying. The paper says HFT has caused spreads to go down because in order to mitigate its effects, third parties are breaking larger trades into smaller ones. But that only reduces the spread on paper by increasing the number of transactions.
I'm not sure what you mean by your second point. Sellers and buyers are always breaking large blocks into small blocks. Moving large blocks of tradable instruments at the best price is one of the fundamental technical problems of trading. The spread captures what at any moment you would need to give to middlemen to transact on the market. Smaller is better no matter what the reason.
The value provided by "middle men", be they robots or crooked traders in pits, is liquidity and price discovery. Trading allows people holding an instrument to unload it more quickly rather than waiting to convince themselves to take a bigger leap on pricing; it's what, in a simplifier Bohr-model kind of way of looking at stock trading, allows you to buy shares in a stock at a market price you can look up online as opposed to waiting for the stars to align. The faster the trading, the lower the spread, the smaller the leap sellers and buyers have to take.
It's worth noting again as always that before automated and program trading, spreads were higher, and the space in the spread between the bid and the ask was basically a license to steal.
Here's a good comment from awhile ago, including a cite to an accessible paper:
http://news.ycombinator.com/item?id=3895208