The order of magnitude chosen to calibrate the metric on doesn't change the prefix magnitude.
1 kJ = 1000J
1 kg = 1000g
Adjusting prefixes puts the units on the order of magnitude where they are easy to write in typical domain problems. For a heat-capacity and heat-transfer problem, I might end up with J/kg or kJ/kg being convenient, while for expressing the specific energy of jet fuel, I'll use MJ/kg because it's 39.
In the case of the kg, I'll bet that 1g was simply too small for the tools of the time to make an accurate model that was reproducible to the desired accuracy. For all I know, the weights were die-cast from the same melt, and the effect of scratches gets reduced as the surface-area/volume ratio goes down.