To try and answer this question I collected four 1H-13C HSQCs each with a different number of scans (1,2,4,8) and t1 increments (512,256,128,64), but with the same total experimental time (11 minutes 34 seconds). The sample I used was 5% ortho-dichlorobenzene in acetone-d6, which shows two aromatic resonances. Shown below is an expansion of the aromatic resonances in the overlaid HSQCs. The spectra are offset so that all the peaks can be clearly seen. The first thing to notice is that the width of the peaks in the 1H dimension is the same in all the spectra, but the width in the 13C dimension increases as the number of slices collected decreases.
Measuring the signal to noise in a row taken through the left peak we find that increasing the number of scans increases the signal to noise, as you might expect, but a doubling of scans does not double the signal to noise. The increase is in fact much less. Measuring the linewidth of the same peak in the 13C dimension we find that the linewidth scales linearly with the number of slices.
|Spectrum||Scans||Slices||1H S:N||13C linewidth (ppm)|
Since linewidth, and thus resolution, scales linearly with the number of slices, whereas signal to noise does not scale linearly with the number of scans, increasing the number of slices is a more effective use of time. In practice, you need a minimum number of scans to detect a signal and a minimum number of slices to obtain sufficient resolution for the experiment to produce useful data. If you have more time available than what is required for the bare minimum number of scans and slices then I recommend increasing the number of slices.