Once the scanner has acquired the data, and performed the FFT, it has the image represented in memory in floating point format.  The output file consists of 16 bit unsigned integers (range 0 to 65535), however.  So before saving to file (and to the database), the floats are multiplied by a scaling factor and converted to integers.  If the scaling factor is too small, the range of intensities in the final image will be all in the bottom few values e.g. in the range 0 to 500 instead of the more appropriate 0 to 5000.  In the former case there would be more discretization noise i.e. if you choose too small a scaling factor, the SNR of your final image will be worse.  The significance depends on what the noise level due to other sources is relative to the discretization noise.  Most post-processing programs probably scale the image anyway, so you might not notice the difference in the end.  For example, FreeSurfer converts the image to an 8 bit/pixel representation and then intensity-normalizes it.  Nevertheless, it's best to scale correctly so as not to add discretization noise to your data.  The "scaled" sequences are set up to automatically use the correct FFT scale factor.