As I understood it, the reason was because of segmented architectures where pointers are stored as segment id + offset, and thus comparison using < would only require comparing the offsets, whereas std::less would require comparing both the offset and segment id to ensure that for two pointers with the same offset A < B == B >= A.

Under such a scheme making < and std::less equivalent would mean introducing a redundant check when comparing pointers with the same array (which I imagine is usually the most common case for using < on two pointers) and potentially inducing a performance penalty if the segment id for one of the pointers is in some processor state register and needs to be moved to a gpr first.