I think we already have more than an inkling of the usefulness of suffering over warning signs which are less burdensome to experience. It can be awfully tempting to override such warning signs when we can.
Imagine a group of hunters who're chasing down a valuable game animal. All the hunters know that the first one to spear it will get a lot of extra respect in the group. One hunter who's an exceptional runner pulls to the head of the group, bearing down on the animal... and breaks a bone in his leg.
In a world where he gets a signal of his body's state, but it's not as distressing as pain is, he's likely to try to power on and bring down the game animal. He might still be the first to get a spear in it, at the cost of serious long term disability, more costly to him than the status is valuable.
The hunter's evolutionary prospects are better in a world where the difficulty in overriding the signal is commensurate with the potential costs of doing so. If attempting to override such signals were not so viscerally unpleasant, we'd probably only be able to make remotely effective tradeoffs on them using System 2 reasoning, and we're very often not in a state to do that when making decisions regarding damage to our own bodies.
In the recent discussions here about the value of animals several people have argued that what matters is "sentience", or the ability to feel. This goes back to at least Bentham with "The question is not, Can they reason? nor, Can they talk? but, Can they suffer?"
Is "can they feel pain" or "can they feel pleasure" really the right question, though? Let's say we research the biological correlates of pleasure until we understand how to make a compact and efficient network of neurons that constantly experiences maximum pleasure. Because we've thrown out nearly everything else a brain does, this has the potential for orders of magnitude more sentience per gram of neurons than anything currently existing. A group of altruists intend to create a "happy neuron farm" of these: is this valuable? How valuable?
(Or say a supervillian is creating a "sad neuron farm". How important is it that we stop them? Does it matter at all?)