[lommelun]

1) Towards the end of a production cycle mostly laborious work is remaining that doesn’t require as much creativity to produce. Not so much programmers working on an engine, but rather art teams working on finishing assets. For this kind of work you do get more output per hour worked, even if people are overworked. Some companies (many companies unfortunately) do plan for this crunch early on.

2) For a multitude of reasons, usually related to biz dev and budgeting, release windows are set way ahead of time. Then a marketing campaign starts months in advance to build up to release. A marketing campaign is extremely costly, and loses much of its effectiveness if it’s interrupted. it might be impossible to build up the same level of attention a second time, and most companies aren’t willing or can’t afford to take that chance.

3) The work remaining that does require creativity, responding to QA demands, porting, redesigns, bug fixing, etc ends up in a crunch by necessity due to 2).

[patrec]

The first part of your answer that sounds intuitively believable, namely that a bunch of chores accumulate by the end of a production cycle and that productivity is not as badly hit, so it might be possible to get more out of people by overworking them (it's not completely obvious though, many of the studies about productivity declines due to overwork I skimmed study tasks that don't seem like they'd obviously suffer more under overwork than art asset creation).

But the other two points are in a way begging the question: I understand that both poor planning and set-in-stone release/marketing windows are endemic in the industry (and that this creates temptation to overwork people to hit a slipped deadline). But if overworking people is generally a net loss in throughput (as a lot of studies seem to suggest), giving in to this temptation ought to be counterproductive and decrease the likelihood of meeting the deadline. Now of course it's not uncommon for people to really persistently, and en-masse do stupid things that turn out to be highly counterproductive -- like seeing a doctor throughout most of human history. On the other hand, given that the practice seems to be both ubiquitous and universally hated you'd think the first major game studio to figure this out would completely crush its competitors in no time. Since this hasn't happened yet, it's tempting to conclude it must provide some advantage.