The simplistic explanation is, if you're doing arbitrage - i.e. "fixing market mispricing", there's only so much arbitrage you can do before you fix the price...
This is of course a completely theoretical proposition, because in reality you don't know what the "fair price" is. You don't even have probabilities, because those are also unobservable, you only see one version of "history".
In practice, what happens is that if you trade "too much", "shit goes wrong". Both of these things require empirical estimation and are easy to get wrong.
The most obvious is the market liquidity, which you can observe at e.g. BitStamp TradeView [1] - there's only so many orders at a given price, so the more you trade, the worse price you get (the average/marginal trade).
No professional of course trades like that, especially not HFTs, but similar problems happen at every scale - you're competing with other traders, they might have better information, there's limited amount of stock in the market, the edge/alpha/expected profit you can earn decays over time as the price moves, if you trade too much you move the market and inform other participants who can then trade against you, ...
[1] https://www.bitstamp.net/market/tradeview/