When Nintendo made the Wii, they purposefully left themselves production constrained.
Tesla is a different beast, but consider this:
Cost per vehicle is BOM + labor per car + overhead per car
BOM drops a little with volume and overhead per car drops fairly linearly with volume. However, labor does not have so easy a trend line because it (typically) gets added in full shifts.
Say one shift fully utilized makes C cars at a labor cost of L/C per car.
Bump production 20%, now a full additional shift is needed (unless you go with irregular scheduling). So you have 2L/1.2C = 1.67L/C, a 67% increase in labor cost per car. Berlin has labor rules which make additional shifts even more disadvantageous so it's >2L.
The 17% reduction in factory overhead per car isn't going to make up for that, especially considering factory overhead is really a constant that gets amortized at the end so there are no savings in volume. There is also the step change of stamping and casting capacity to deal with.
To optimize margin, Tesla wants to get demand aligned with maximum output of the current shift capacity. A small increase in demand beyond that works against them. Either wait time starts increasing or they bump up pricing to reduce demand. Only when there is a large amount of excess demand does shift adding make sense. Cybertruck's backlog definitely supports multiple shifts.