I suppose the short answer is "tradition." Many western languages work this way - where a verb-expression can be like the object of the verb - so it's natural that Esperanto works the same way. Nouns, pronouns, and adjectives are marked for accusative when they're direct objects, but verbs are not.
If we want a longer answer, we first need to ask what we would gain by allowing verb expressions to be marked as direct objects. Just as you ask why this is not done, one could just as easily ask why it should be done. I don't see any good reason. When you consider that it's not just verb expressions, but whole phrases which can be the object (Mi scias, ke vi havas bananon), it just seems like an unnecessary complication.
Another part of the longer answer is to remember that the direct object is that which receives the action of the verb. IF we mark verbs as receiving the action of the verb, how would we know which verb it's receiving the action of. Often these verbs have their own objects (Ŝi ŝatas manĝi pomojn.) Again, marking verbs as the object of one verb but not another seems like an unnecessary complication.