Video games don't work well with this model because in most video games a) information is perfect and b) results are deterministic. You don't have to update your understanding of the world constantly because your knowledge of the world is more or less complete.
In John Boyd's world, you couldn't "pre-compile" any decisions because he was working in the real world, not a game.
Neither of these are true in the strategy games the parent comment refers to. They mention playing "in real time", and I interpret that to mean real-time strategy games. In those games, you generally have:
1) Fog of war. You can only see the play area near where you have units. Enemy movements outside of that region are hidden from you. So you are constantly discovering new information as your unit moves around, and having to deal with where you thought the enemy was being wrong.
2) Randomized damage. Most attacks don't do a fixed amount of damage and instead do a certain amount of "rolling the dice". You know on average how much damage X will do against Y, but not the results of any single hit until after it has happened.
The reality is that the vast majority of real-life situations are hybrids[0]---some things are constant, others are new. The math we do changes, but math doesn't change. Boyd's OODA period may have been shorter than that of enemy combatants, but he still had to go to training. What is training if not pre-compiled thought? And importantly: in fluid environments where novel things are likely to arise, having a solid grasp on things that are "settled" makes it easier to integrate new phenomena.
Take fencing, for example. At low levels, people often try to make up for poor technique with speed and strength. Like you point out, this is a mistake. But at the highest levels, where everyone’s technique is basically flawless, the winner is the fencer whose thinking is a step or two ahead.
That said I’m really talking about situations where you have an active opponent. With more static problems yeah, being correct is more important than staying on your toes.
Or at least, I'm skeptical that that's the best way to frame a lot of problems.
A better way, IMO is to frame it in terms of mental models.
Some mental models are more efficient to think in, just as some programming languages are more efficient to write in. Many domains are amenable to study beforehand, allowing you to figure out the right model to think about the domain, effectively pre-compiling the Orient and Decide steps.
This can be a big deal for two reasons. The more boring one is that pre-doing anything will probably give speed gains. The much more interesting one is that you can get qualitatively different Orient/Decide steps. A buttload of contemplation, condensed into a few maxims, is usually worth a lot more than thinking slightly (or even two or ten times) faster in a critical situation.