It's different in the way that the bump maps have been created. Traditionally for years, bump maps were created by an artist scanning or "painting" a height-map/graymap and converting it into a bump map. Basically, to give some random rough texture to the surface. Either it would be hand-aligned to the existing geometry, or worse, just thrown onto the model without regard (e.g. the orange-peel bump texture technique)
PolyBumps (or whatever you want to call them) are created from high resolution geometry. The normals in the normal map represent normals in the actual geometry being exported and decimated, not some artibrary "bumpy" normal texture that an artist hacked up in photoshop.
All 3D is done with vectors and oodles of dot products, so there is "nothing new under the sun" eh? But the devil is in the details of how you use it.