The actual reason relates to the time spent monitoring the ions of interest in the two modes. A SCAN mode spectrum typically requires about 1 second to obtain. If one scans from 50-550 M/z, then the dwell time on one mass is about 2 ms. One is monitoring each ion of interest for only 0.2% of the total scan time and data points are spaced at 1 second intervals.

In SIM mode, the fact that you spend the entire time monitoring only the ions of interest means you have longer time to accumulate data for those ions and hence get a larger signal. At the same time, the longer monitoring time provided more opportunity for noise to cancel. The S/N improvement should be related to the relative time spent integrating each ion.

Another problem in fast GC-MS analysis is that peaks are often only a few seconds wide. One needs at least 8 data points to accurately define a gaussian peak. Thus a SIM mode scan can be set up to acquire more data points across a peak and provide a more accurate profile.