To understand convexity, all you need to do is a simple cost benefit analysis.
Spaced repetition algorithms create an asymmetry between the gains from learning and the losses in time by optimally spacing repetitions. They minimize the time you need to spend to acquire and retain knowledge. In this way they "cap the downside" of learning by minimizing the resources spent to acquire gains.
In terms of the potential gains from using a spaced repetition system, I argue that the potential value of using SRS varies dramatically based on the learning domain. I divide learning domains into "quality domains" and "quantity domains" based on the shape of the graph of the payoff function.
Quality domains tend to be creative domains like science and programming where the gains from each additional flashcard you learn are potentially unbounded. The fact that the potential value from learning in these domains is unbounded creates a convex shaped payoff graph.
For example, think about scientific research where a single serendipitous association could generate a new breakthrough. By using SRS in scientific domains, you spend little time acquiring and retaining knowledge compared to the potential gains. I call these domains "quality domains" because the number of flashcards you make is less critical than the quality of the concepts and connections you encode, so you can maximize your chances of creating novel ideas that can be applied in the real world.
Other domains like language learning I would describe as "quantity domains". Success for foreign language learners is defined as having a comprehensive vocabulary which requires a large collection of items. The potential gains from each additional flashcard are bounded and decrease with each flashcard as the learner acquires words that occur less and less frequently in real life. There is little to zero chance that a card in a quantity domain could create a serendipitous connection that results in a new scientific breakthrough, a startup idea or a new invention, so the graph of the payoff function is not convex.
The use of spaced repetition caps the cost of learning by minimizing the time spent to acquire and retain knowledge. If you decide to invest your time using the spaced repetition system into increasing your knowledge in creative domains, then you are enaging in a process with a large asymmetry between the cost of learning and the unbounded potential gains.
By contrast, quantity domains benefit from the capped costs of learning by optimally spacing reviews, but they don't really have convex payoffs, because the probability of you generating life-changing creative associations is negligible.