Mean interaction contrast has been shown to be a beneficial statistic for testing among various mental architectures (e.g., Sternberg, 1969; Schweickert & Townsend, 1989; Townsend, 1984). In particular, within the context of parallel vs serial processing, a negative contrast (i.e., underadditivity) is supportive of parallel processing whereas processing. The present theoretical note investigates a speculation of Egeth and Dagenbach (1991) that the mean interaction contrast is largest when the parallel processes are each deterministic, that is possessing zero variance. We verify their conjecture but show that the mean interaction contrast is not necessarily monotonic decreasing with variance. Experimental implications are discussed.