If you have large images (>5000 columns) then it is advantagous to have images with pyramids created. For small images there is little or no advantage.
The pyramids (such as .OVR file, .RRD; internal overviews in TIF files or MrSID etc) are used by ArcGIS (and in Mosaic Datasets) to access imagery at smaller scales. Consider the case of a large image with say 20,000 columns. If pyramids are created they would have multiple levels with 10000, 5000, 2500, 1250, 625 columns. These multiple levels of pyramids would be used making access faster at the smaller scales. The screen area covered by the pyramids gets smaller and smaller though and a point is reached when covering the screen would require opening too many files. By default when rasters are added the LoPS value is set based on the level at which the Num Cols goes <1500. In the above cased 2500. When generating overivews these would then get created at a scales equivalent to about 2x this scale. The result of this is that the overview are created at scales when the number of colums in the pyramids woudl go < about 1500 (or when too many images would generally be needed). The use of pyramids therefore increases the scale at which overviews are created. For overlapping imagery this has the advantage of increasing the scales at which the mosaic methods work and reducing the number of overviews. For applications where there are only sets of tiled images, or if the images are small then there is no advantage of having pyramids.
