The underexplored value of early segmentation was the discretionary segment level permissions enforced by hardware.
Years ago I prototyped a system that had filesystem permission support at the segment level. The idea was you could have a secure dynamic library for, say, manipulating the passwd file (you can tell how long ago that was). You could call into it if you had the execute bit set appropriately, even if you didn't have the read bit set, so you couldn't read the memory but could call into it at the allowed locations (i.e. PLT was x only).
However it was clear everyone wanted to get rid of the segment support, so that idea never went anywhere.
They made a decent go at it again in 16 and 32 bit protected mode. The GDT and LDT along with task gates were intended to be used as an hardware object capability system like the iAPX 432's.
I'd hope that anyone at Intel with said nightmares would have read this paper by now (wherein Bob Colwell, et al, argue that the 432 could have been faster with some minor fixes, and competitive with contemporary CPUs with some additional larger modifications).
Having said that, I wouldn't be surprised if some form of segmentation became popular again.