Reflection seismic data are nearly always adequately sampled in time. Four millisecond sampling is typical, giving a temporal Nyquist of 125 Hz, more than adequate since the signal tends to die out somewhere in the 60-100 Hz range. In contrast, seismic data is often inadequately sampled along various spatial axes. If there is more signal, improving temporal sampling is easy. Improving spatial sampling is expensive, because it requires more equipment, time and/or manpower in the field. Seismic acquisition is expensive. Moreover, reflection surveys tend to be designed with the goal of maximizing poststack or image resolution for a fixed cost (fixed cost fixed number of traces), at the expense of fold. This represents the choice to spend money on well-sampled geology, in order to get more precise subsurface information, rather than on a well-sampled wavefield which would be easier to process.
All of the multichannel steps in seismic data processing depend to some degree on adequately sampled data. Inputting spatially aliased data can produce results with various sorts of artifacts and poor spatial resolution. Marfurt et al. 1996 give examples of artifacts that appear in Radon transforms of insufficiently sampled data. Spitz 1991 shows differences between migration results on well sampled and poorly sampled data.