Haven't stood on this soapbox for a while...

Doubling the Shadow Size does result in a somewhat sharper shadow edge resolution, yet increases the memory demands four-fold for each light. Four times larger =16 times the RAM demands -- and so on.

Why tie up memory unnecessarily, and increase render times, with larger shadowmaps when you can usually do the same for free?

Changing the Zoom level or Scale on the light's ShadowCam, can almost always yield the same change in shadow edge detail, and also permits unlimited scaling for really huge variability in shadow edge feathering, without ANY additional memory demands.

Unless tweaked, the view from each shadowcam, from which the shadowmaps are generated, usually encompasses vast tracts of space in your scene that are not visible in the actual render (such as the entire Ground plane mentioned above), meaning that you are wasting shadowmap pixels, and render time, generating shadows for stuff that will never be seen.

As SamTherapy said above 1024, or certainly no larger than your actual render, is plenty big. For sharper shadow edges, simply zoom the shadowcam view in. For softer, more feathered edges zoom out.

Same shadow map size, same memory usage, same render time.

Who says there aint no such thing as a free lunch?

(with apologies to RAH)