It doesn't seem to be a necessary one either, given the existence of counterexamples. Eg. Pierre Omidyar started E-bay while keeping his day job and adamantly refused to work more than 8 hours/day on it even after going full-time. One of the reasons he stepped back from a management position was that he didn't culturally fit in with E-bay after hiring a few work-through-the-night cowboy coders.
The article is talking about something completely different: putting yourself in a state of constant sleep-deprivation through working ridiculous hours. And the author is right: that's not half as effective as people claim.
Fortunately, that can be one of multiple people working in shifts so nobody has to be up for 36 hours at a time.
A better analogy might be playing while hurt. You're clearly not at your full potential, but you are making progress.
But I'd argue that I'd rather have people on my team that aren't hurt: it is detrimental to performance in a similar way and acknowledges the commitment.
And it is also usually a bad idea.
But I still disagree. Anecdotally, I've found that coding for 6 hours a days gets more features implemented and bugs fixed than coding for 14 hours a day.
The loss of mental acuity, clarity and creativity (again, this is for me, and anecdotal) is simply too high a price to pay.
Sure, in the long term. But when you have a big demo or a customer deadline coming up in 2 days, or a server crashing right now, it often pays to keep working past 6 hours.
I hold up the Schilling's bloody sock as Exhibit A:
But the playing while injured metaphor has some parallels as well. Not only is this a badge of honor, but it's flatly expected of you once you get past a certain level of competition. This is generally good and right.
However, past a certain level of injury, you're hurting both yourself and the team, and you're better off taking time to heal.
My personal experience with sleep deprivation is that if I go too long and too hard, I am doing less work than if I had rested, even in absolute (and not per hour) terms. And DHH is 100% correct that as most people get tired, their solutions to problems get worse -- that's definitely true for me.
I've tried various levels of sleep over the years, from a solid 8 hours a night to so little sleep that I would literally hallucinate during the day. For me, except in short (2-3 day) bursts, at least 6-7 hours of sleep a night is most effective for my ongoing productivity. Any less than that, and I would have gotten more done net if I had slept.
Having said that, everyone is different, and if you're one of those people who can get 2 hours of sleep on an ongoing basis and still be highly productive, then cool.
Part of it's just the adrenaline, too. And if you have friends goading you on, an allnighter can be pretty fun.
The end result is probably only about 20-25 hours work in two days - i.e. not much more than normal working days. The advantage isn't in the total hours, it's in having the same problems rattling around in your head for a long time.
Likewise for startups. Sleep deprivation is not a sufficient condition for winning, but empirically it seems to be a necessary one. Not long-term, obviously, but during certain peak times. Even at YC there are times when we have to work so hard that we sleep less. On interview weekends, for example.
So the reason people who work hard (not just in startups, but in all fields) treat sleep deprivation as a badge of honor is that it's a sign that one has the necessary commitment not to cave when there are spikes of work.