According to an aptitude-treatment interaction experiment (Lehmann et al., "Metacognition and Learning, 11," 89-105, 2016, N = 47, published in "Metacognition and Learning"), perceptually disfluent texts facilitated retention and comprehension performance (but not transfer performance) only for learners with higher working memory capacity (WMC). No effects of WMC for a fluent text were found (albeit theoretically, fluency may be more advantageous for learners with lower WMC). The findings of our (pre-registered) direct replication (supervised online sample of N = 96) show a substantial deviation from the original results: In contrast to the interaction effect (disfluency and WMC) of the primary study, we obtained null results for disfluency, WMC, and their interaction for all learning outcomes. Our replication data are not indicative of WMC as a boundary condition moderating the disfluency effect on learning. We discuss discrepancies in the results of the primary study and our direct replication regarding particular methodological and analytical decisions, questioning the robustness and generalizability of Lehman et al.'s results beyond their primary study.