I think you are right that things will improve, but for exactly the wrong reason. Truly efficient markets are the toy problems of economic modeling, not typically achievable in real problem domains. Perhaps counter intuitively, I think we will gain efficiency by getting better at modeling the inefficiencies... but the end result of this is no more a "pure" (in your sense) market than a modern particle model consists of Aristotelian atoms.