On the other hand, the formal language of category theory should be learned, and used, at some point. I have seen several interesting papers written by very good mathematicians, containing theorems with statements like "It is the same to give a regular thingamabob over $X$, and a von Neuman whatchamacallit with a seminormal connection over $X'$". What these statements usually mean, is that there is an equivalence of the category of regular thingamabobs over $X$ and that of von Neuman whatchamacallits with a seminormal connection over $X'$; but they could also simply mean that there is a bijection of isomorphism classes, and to know which is true you have to study the proof. This means, I suppose, that the authors, who must have seen the language of category theory at a certain point, have not interiorized it, and don't have a feeling for when its use is appropriate.