You're right that in a vacuum the job is to make the max EV decision at all times, be it a $0.01 better EV. However we do not live in a vacuum, we can't put in the necessary volume for ultra-high variance plays to average out, we have bankroll management considerations, time constraints and short/medium term goals (such as reaching the next stakes as fast and surely as possible within these constraints, which in itself may be more +EV than a better win rate at the lower stakes, by the way).
Let me try to explain this with an example (OP, sorry for hijack). Let's say two players play a modest 10,000 hands per month, or 120,000 hands per year. Look at the two graphs below. The first one simulates 10 possible winning graphs over 120,000 hands of a 6bb/100 winning player who plays a style with a 100bb standard deviation (a mathematical measure of variance). The second one simulates some possible winnings over 120,000 hands of a 5bb/100 winning player who plays a lower variance style with a 80bb standard deviation. Granted, the average winnings of the 6bb/100 player are higher (obviously). However, because of his high variance style, if he runs bad, he could very well end the year a slight looser or breaking even. The 5bb/100 player, even if he runs bad, will end the year at +2000bb, which is 20 buy-ins, which is probably enough to play the higher stakes next season.
See what variance can do to you? In the real world, there is definitely a trade off to make between winrate and variance. You choose the trade off that suits you best. Investors in stocks and bonds make these exact same choices and trade offs every day. And the variance we experience every day in poker would make most of them run out of the room screaming.