Despite the fact that models usually show that education is good for national income, the empirical evidence is mixed. The typical reason provided is that there are inefficiencies in the provision of education, or that raising the funds for public education is distortionary.
Rossana Patrón focuses on the latter using a growth model with an informal sector where households allocate their time to education, formal and informal work. Formal labor income is taxed to finance public education, along with indirect taxes in the formal sector. The economy may end up on the wrong side of the Laffer curve because any increase in the tax rate discourages labor, and in particular formal labor. Thus a government should only raise taxes if the distortionary effect is swamped out by the effect on human capital accumulation.
Now this is all very intuitive, and we would not really have needed a complex model to figure this out. Where such a model becomes really useful is that it allows to quantify things, in particular when there are ambiguities such as the one above. With realistic parametrization, it may turn out that we are far from an ambiguity and that raising taxes for education is a no-brainer (or not). But theory that tells you rather intuitive results and that anything can happen somewhere in the parameter domain is not useful.