The SOB-CS computational model of working memory explains findings from simple and complex span experiments. According to the model’s mechanism of interference by superposition, high similarity between memory items and subsequently processed distractors is beneficial because the more a distractor is similar to an item, the more they share similar units, leading to less distortion of the memory item. When time allows, SOB-CS removes interfering distractors from memory by unbinding them from their context. The combination of these two mechanisms leads to the prediction that when free time is long enough to remove the distractors entirely, similarity between items and distractors should no longer be beneficial to memory performance. The aim of our study was to test this prediction with humans and simulations. Experimental data disconfirmed the prediction. Simulations showed that SOB-CS overestimated the strength of removal which has to be much lower than expected to account for experimental data.