Holly Dunsworth's recent paper and very fine posts here on MT ( the lead-in, the story, what it means for your pregnancy, and cultural and philosophical implications) skillfully discussed important issues about the kinds of explanations people offer for various of our traits. One of the most subtle, and important, was her explanation of the fact that there can be variation among individuals around a species' characteristic trait (which she discusses here).
In this case, as she pointed out, not all women have the same gestation length, there is variation in metabolism within species, and so on. Yet, there are general characteristics of a given species that differ from the similar trait in other species. In other words, the trait is not rigidly fixed, and variation within species and species differences can both be consistent with the same general accounting for the trait.
One way to account for this is to consider the genetic basis of what we call metabolism. Metabolism, like many if not most traits, is the result of many different processes taking place, and these are brought about by the action of many different genes (often hundreds of them), reacting to conditions in the environment. In this case, those conditions involve diet, type and level of activity, perhaps climate and more.
Different environmental experiences can add variation to a trait like metabolic activity and so on, modifying the exact trait value--like days of gestation for an individual fetus. Random effects (chance) also almost always plays a role, since no two humans are identical. Similarly, all the contributing genes, and the regulation of their timing and level of expression, are subject to mutation and vary among individuals.
Successful gestation and maternal survival and infant care can be affected by this genetic variation and for obvious reasons the selective effects can be brutally stark. Maternal or fetal mortality are huge potential fitness effects. So there is no doubt that gestation and delivery can be under strong natural selection. However, that doesn't mean selection favoring one precise direction or trait, or one gene.
Most selection will basically strongly affect women or fetuses who do not 'behave' successfully enough to be born. But each instance will involve different genotypes, and the selection affecting any given variant in a given gene will likely be very small. The distribution of the trait in the population--such as hip-width or gestation length--will have some mean (average) value and some amount of variation. These can change gradually over time, or tend to hover around some rather stable values.
If selection occurs for some other functions that may involve the trait (that is, something about the skeleton or diet and metabolism) distribution of the trait can change. If this were to happen differently in isolated populations, they will diverge and eventually become different species. Even just chance meandering of the trait values could leave them different between species.
As a result, each species will have its own mean trait value, while individuals within the species can vary. This is, I think, just the pattern Holly was describing in her recent post. It is largely the consequence of causal complexity. This is not a yes/no, Mendelian pea-like kind of trait in which there are only 2 or 3 states in the population, one of which is favored and the other lethal, so that different species evolve to have only one or the other variant. It's the result of many different contributing factors.
And this, naturally, ties into the whole issue of the complex nature of genetics underlying traits that, like the ones considered here, are not just simple. That's why it is difficult to identify 'the' gene or a few genes 'for' the trait, because that's not how the trait is produced. And that is the problem with GWAS or other attempts to simplify the basis of what we are or what happens to us.