load-graph.py in multithreaded mode will find slightly different number of unique kmers. This is being investigated in https://github.com/dib-lab/khmer/issues/1248
load-graph.py