Acoustic and contextual cues to linguistic categories tend to be temporally distributed across the speech signal. Optimal cue integration thus requires maintenance of subcategorical information over time. At the same time, previous work suggests that finite sensory memory or processing capacity strongly limits how much subcategorical information can be maintained (or for how long). We argue that previous work might have over-interpreted the role of these limitations. In two perception experiments, we find no limit in the ability to maintain subcategorical information. We also find that maintenance seems to be the default, neither limited to perceptually particularly ambiguous signals, nor a learned strategy specific to our experiment. In contrast, listeners' decision for how long to delay categorization, we find, is a function of perceptual ambiguity. It is therefore crucial to distinguish between in-principle abilities (even when they reflect default processing), and decisions made within the bounds of those abilities.