Hi Frédéric,
always impressed by your work on this topics 
I will definitely have a look at the git repository.
2 quick questions meanwhile:
1- the first open_mfdataset, is it loading netcdf files ?
2- I also tried some netcdf / zarr conversion on my side so your publication is very useful. By the way, do you have any advice on the performance of loading of many netcdf files BEFORE building the zarr dataset ? I found open_mfdataset quite under-performing on a large number of files but I did not found a practical alternative (see a detailed question in the pangeo forum here)