It is common to suppose that if we can identify a change that would increase fitness, then evolution will make it happen. Cost/benefit trade-offs, although important, are not the only impediment. You also need (a) the opportunity (which is why apes, having entirely lost their 5th appendage, are unlikely now to evolve a prehensile tail) and (b) the ability to benefit sufficiently from an achievable incremental change (which is why seals beaching on predator-free islands have so far failed to evolve into land predators - they have some land locomotion already, but somewhat improving it won't be sufficient to make any material improvement to their land predation potential, they'd need a large improvement to do that).

Fitness and cost/benefit is always relative to the conditions, which includes who else is out there. Since that is always changing, to say something evolves "until" needs to be carefully interpreted. The parasite/host evolution battle seems to proceed perpetually.

So maybe flies don't learn more in the wild because they don't have the opportunity. Though probably the author has considered this possibility and discarded. Nonetheless it can be real case, at least in theory. For example India's achievement in sport is very poor in comparison to the rest of the world, though one would think the possibilities just as lucrative for one who did succeed, but on the whole its population just doesn't get the opportunity.

Another interesting analogy comes from the neural network programmes that play games like backgammon. These programmes learn how to play the game well (better than humans, in the case of backgammon) by playing it lots of times and seeing what happens, and thereby improving their situation analysis algorithm. But what happens is that they can fail to encounter the opportunity to learn certain things, unless they are forced to learn. For example, if they take an early erroneous view from learning so far that certain situations are bad, this can act as an impediment every to getting into those situations - you won't get their at random unless you cooperate. But in fact some of those situations are good. What has happened is that we have had to "tutor" these programmes by forcing them into those situations, and then they learn that they are good.

Animals with parental care probably have more opportunity to learn - their young can be forced into the learning situation that they wouldn't encounter on their own.