The standard wisdom in Celestia is generally to favor low resolution meshes combined with high-resolution texture maps, in particular, normalmaps.
ie. sacrifice load time
for better FPS.
This usually works really well when viewing from a distance (if the texture and normalmaps are accurate, and well matched).
Up close to a mesh however, there are some benefits with a higher resolution mesh, as this series of the same view of the rim of Shackleton Crater, with a range of different mesh resolutions and textures may demonstrate.
Each is shown in wire-frame mode on the left, and with it's texture on the right (click for large pictures as usual).
All the textures are either 4K or 2K, which exceeds the resolution of the mesh in most or all cases.
The first 4 use a texture derived from a LOLA shaded relief map.
Firstly, a mesh which covers the 10 degree range from 80 degrees south to the South Pole.
The source IMG file has a resolution of 40m/pix by 40m/pix (15,200 x 15,200 samples).
I have down sampled this by a factor of 10 to give a final mesh of 1520x1520 vertices (or a net resolution of 400m/vertex).
Next, a mesh which covers the 5 degree range from 85 degrees south to the South Pole.
The IMG again has a raw resolution of 40m/pix by 40m/pix, but covers half the range of the first (7584 x 7584 samples).
Again, I have down sampled by a factor of 10 to give a final mesh of ~ 758x758 vertices (at the same resolution of 400m/vertex).
I think you'd agree that there's not a huge difference to be seen between the 2, either in wire-frame or texture mode.
(Not really that surprising given that their effective resolution is the same.)
Next, we halve the range once again to the 2.5 degrees between 87.5 and 90 degrees south.
The IMG file in this case has a raw resolution of 20m/pix by 20m/pix, (7584 x 7584 samples).
Again, down sampling by a factor of 10 gives a final mesh of 758x758 vertices (but this time with a resolution of 200m/vertex).
Now, in the wireframe view of things, we are just beginning to see hints of the terracing of the walls of Shackleton Crater, and the surrounding area, although this detail is largely hidden when viewed with the texture...
Next, using the same IMG file as input, we sample at a higher resolution by choosing a sample factor of 3, resulting in a mesh of ~ 2516 x 2516 vertices.
(or a net resolution of 60m/vertex).
Now the terracing is quite obvious and detailed in wire-frame mode (but any advantage is again not realized by the use a texture map which doesn't match well with the IMG file.
The final 2 examples use the same 2 meshes as the previous 2, but in these cases, the texture used is one that is generated from the IMG file, and so it matches perfectly with the features of the mesh.
The 200m/px version:
and, the 60m/pix version:
Clearly in the case of these last 2, the accuracy and quality of the texture would allow the use of the lower resolution mesh without significant loss of detail (at least from reasonable distance anyway).