With all due respect to the master Staale and this discovery, all this only addresses half the factors without taking into consideration the settings of the ShadowCams.

Depending on how much of the actual render is covered in the viewing angle of the each light's ShadowCam, (the view from which each shadowmap is actually generated) some, or even most of the pixels in the shadowmap may fall outside the view of the rendering camera and may not be utilized in the actual render.

A 8000x8000 pixel shadowmap generated from a shadowcam that is viewing a field that is 8 times larger than the scene being rendered will yield the same shadows as a 1000x1000 pixel map that is generated from a shadowcam that is viewing only the portion of the scene being rendered. The same number of pixels will be used and the rest disregarded.

The point being that maps of vastly different sizes can produce identical results depending on the shadowcam field of view. If you only adjust the map size without addressing the shadowcams' views, you may be wasting huge amouts of memory & render time on calculating shadowmap pixels that will not affect the render.

(gotta do a tute with pics.....one of these days.)