Somewhat on topic: I would guess there's some value in having a separate base pointer for global image data versus code. So that data across images could be packed more tightly rather than being interleaved. Though maybe the fdpic linkage model is good enough for that purpose.

Also it seems like it could be valuable to associate some access rights based on code address, like apple did with their arm chips. It looks like they implemented a mini hypervisor in their kernel to hold the privilege of changing PTEs. One could similarly partition access to the global memory map and page cache to separate high privileged code.