You can try citing
S. A. Huettel and G. McCarthy. The effects of single-trial averaging upon the spatial extent of fMRI activation. Neuroreport 12 (11):2411-2416, 2001.
In this paper thy have a graph of stdev vs #trials averaged. The 'knee' of the graph is about 15 trials. Of course, this is for their particular experiment, but it does indicate that 18 trials is not necessarily a bad thing.