P.S. And that speed (~700 ms) is both for chuncked and unchunked data so I decided that is was "slow" for the chunked file since it is considered more optimized for this read pattern.
netcdf-java