In the first frame most of the work is done closing the compressor (somethings not quite right, either in user land or the library implementation). In the second frame (a different library) there's no such close overhead. 5X faster reads. Another win for performance profiling.
