Outside my depth but I'll give it a stab. Identify what data is important, (is the full 42Tb needed?). Can the data be split into easier to handle chunks?
If it is, then I personally do an initial sneakernet to get the fist set of data over. Then mirror different on a regular basis.