While some difficult learning conditions can improve learning, the findings regarding the contribution to learning of disfluent, hard-to-read text materials have been inconsistent. We identified test delay and disfluency manipulations as factors potentially contributing to these discrepancies. We tested students' immediate and delayed memory performance (2 weeks later) on a course text that was presented between-subjects (N = 134) either as perceptually disfluent with a hard-to-read-font, as lexically disfluent with 20% scrambled letters, or in its original format. By distinguishing between short-term and long-term learning, our expectations were supported; an illegible font reduced forgetting, thereby producing delayed memory benefits. We also tested whether lexical disfluency would have similar memory effects as perceptual disfluency, as the meta-cognitive perspective suggests, or whether different disfluency manipulations would have different memory effects, as ideas from a contextualized framework on desirable difficulties suggests. The findings supported the latter. The results are discussed regarding the generalizability of the disfluency effects and the implications for when disfluency is desirable.