Sequences of auditory objects such as one-syllable words or brief sounds disrupt serial recall of visually presented targets even when the auditory objects are completely irrelevant for the task at hand. The "token set size effect" is a label for the claim that disruption increases only when moving from a 1-token distractor sequence (e.g., "AAAAAAAA") to a token set size of 2 (e.g., "ABABABAB") but remains constant when moving from a token set size of 2 to a larger token set size (e.g., "ABCABCAB" or "DAGCFBEH"). Here we show that this claim was incorrect and based on experiments with insufficient statistical power. With sufficient statistical power it can be shown that disruption increases not only when the distractor token set size increases from 1 to 2, but also when it increases from two to eight one-syllable words (Experiment 1) and brief instrumental sounds (Experiment 2). These findings have implications for theories of auditory distraction which differ in their predictions about whether the distractor-induced performance decrement should (a) only be determined by acoustic differences between immediately adjacent distractor tokens (duplex-mechanism account) or (b) gradually increase as a function of the variability in the distractor set (attentional account). The present data are inconsistent with the duplex-mechanism account and support the attentional account.