Large Language Models Imitate Logical Reasoning, but at what Cost?