In a previous experiment [McCloy and Lee, 2013, J. Acoust. Soc. Am.134, 4230 (2013)], we reported an asymmetry between phonetic and semantic detection tasks with respect to spatially separated vs. spatially adjacent attended word streams: phonetic tasks showed high false alarm rates when to-be-attended streams were spatially divided, while semantic tasks did not. In this experiment, we manipulate the difficulty of the semantic task to investigate the effect of cognitive load on task performance. We compare trials in which the two to-be-attended streams comprise words drawn from either one or two semantic categories, and from categories with either a small or large number of words. Effects of these manipulations on target hit rate and false alarm rate are discussed in relation to previous work.