merge_grids.py - 3D field merging is slow

The speed of merging 3D fields dropped dramatically when I switched to compressed input netCDF files. Merging 3D output (only ql and qr) from a 256x256x252 run now takes 24h on lisa.

Is there some way to speed up the copy operation, maybe making sure that data is read from the source files in large blocks?
