Since this operation works on individual pixels (rather than blocks), it is capable of creating predictions that are not possible using block-based approaches.
This choice, while costing an additional bit per macroblock, can improve prediction quality and therefore reduce residual.
Because the transforms used in global motion compensation are only added to the encoding stream when used, they do not have a constant bitrate overhead.
GMC failed to meet expectations of dramatic improvements in motion compensation, and as a result it was omitted from the H.264/MPEG-4 AVC specification - designed as a successor to MPEG-4 ASP.
[1] GMC also represents a large computational cost during encoding which yields relatively minor quality improvements.