Yes this is something to be concerned about.
Thanks for the report. This has me quite baffled. I'm going to checkout the version from Oct and see if I can reproduce issues with w15o. My thinking is that different results from different runs is pointing to a data race somehow. The non-determinism from the threading can affect file sizes and the partials chosen, but if the contents of slice_counts.csv is different for different runs of the same parameters then something's gone wrong.
Update: I'm seeing differences in slice_counts.csv from run to run. They're small differences and they sometimes take a few hours to show up, but they're there. It may take me some time to figure out what's gone wrong, but right now no negative results from lsss can be trusted.