That is the theory, re reproject-on-the-fly, unless minor adjustments aren't accounted for
As for the offset, I am not sure whether 'cartographic' grade adhere to the passing through cell-centers rule-ish either.
There is also the 'snap raster' and cell size issues when trying to compare two interpolated surfaces and hence their derived contours.
The only way you can definitively examine this is with constructed surfaces which I have dabble with a bit for algorithm testing
Terrain creation... diamond square