The equation is simple about the targeting of the warhead.
I think many of the previous scenario was simply unrealistic .
Say the ballistic missile use radar to find the target.
Say the missile radar has unit diameter, the shore targeting radar has 10 unit diameter.
IT means if the shore radar has 1000 km range , and say 20 km precision at that distance ( proportional to the nebo-m capability) then the warhead radar from 100 km will have inferior capability compared to the shore radar, due to the low energy.
It will get comparable/better picture around 50-60 km from the target than the shore radar, but still with 10 km error.
ok, same math :
missile max acceleration (hypothetic ) is "a"=100 g = 1000 km/sec^2
Missile speed "v"=3000 m/sec
missile turning radius : r=v^2/a=3000*3000/1000=9000 m
So, if a missile with this parameters start to turn at 9 km altitude then it can hit anything on the surface in an 18km diameter circle.
Means the most likely strategy is to cue the missile by the shore radar, it descend to say 20 km altitude with full speed, at that altitude it blow the molten salt batteries and push all energy into its radar ,search 3 sec and manoeuvre itself toward the target with full g.
During the final 10km its track the target, and relay back to the fellow warheads/shore installation the data.
The shore correct the trajectory of the other warheads, and depending on the picture generated by the descending warheads launch more and more missile .
Using MIRV decrease the radar diameter and available energy, so doesn't make sense.