The late Dave Philips wrote that the GS is indeed set up so as to bend / level off at the runway end, instead of continuing in a straight line all the way to the tarmac.
In CAT3 ops this is fine because the autopilot switches to the RADALT at 150ft and then flies an internally predefined profile until wheel spin-up is detected.
Today, at Le Touquet / LFAT, I noticed that the GS definitely started to “bend upwards” at about 300ft. The descent continued ok but at a reduced VS.
If I understand correctly, this is not really an intentional design but rather an artifact of ground proximity. Normally, GS signals reflected off the ground are disregarded by the ILS receiver because the propagation path length difference gets them out of phase, but at low heights this difference would be small enough for the reflected signal to be seen as well, essentially making the GS horizontal. However, this is dependent on both terrain and the type of GS transmitter antenna system. There is some interesting reading on GS antennas here.
I think since LFAT is only CAT1 ILS, they have no need to make it work at all below their published DH. At other places where I tested the ILS below the DH (in VMC obviously) it took me much lower – e.g. Bournemouth, Southend, Norwich, Exeter, even Lydd.
The radio/ electromagnetic waves that produce the vertical signalling on an ILS can be easily corrupted below eg CAT1 200ft DH by ground effects, panels, or whatever else you might find on an airfield.
