The lower efficiency for conjunctions is due to noise and not serial visual attention

Models of human visual processing start with an initial stage with parallel independent processing of different physical attributes or features (e.g. color, orientation, motion). A second stage in these models is a temporally serial mechanism (visual attention) that combines or binds information across feature dimensions (Treisman & Gelade, 1980).  Evidence of this serial mechanism comes from a fundamental finding in visual search of a target among a set of distractors (feature vs. conjunction dichotomy). We have conducted visual search accuracy experiments that carefully control for low-level effects: target/distractor physical similarity, element eccentricity, and eye movements. The results show that the larger set-size effects in visual search accuracy for briefly flashed conjunction displays are quantitatively predicted by a simple noisy parallel model that takes into account that each feature dimension is processed independently with inherent neural noise and assumes that information is combined linearly across feature dimensions. This model is an extension of the classical Signal Detection Theory model (SDT, Green & Swets) and successfully used by John Palmer, Preeti Verghese, Misha Pavel, Nagy and others to predict visual search accuracyin single feature displays. In this model visual attention selects the relevant (noisy) sources of information (cued locations) and allows the observer to ignore irrelevant (noisy) sources of information.  
The human results are not predicted by a temporally serial mechanism (Bergen & Julesz, 1983; Treisman & Gelade, 1980) or by a hybrid model with temporally serial and noisy processing.  The results do not support the idea that a temporally serial mechanism, visual attention, binds information across feature dimensions

(Reference: Eckstein, Psychological Science, 1998; Eckstein et al.,2000 Perception & Psychophysics).