I haven't looked at the code, but I suspect that those flakes have hit another snowflake and are now considered "fallen" snow. And for fallen snow there's probably an optimization that only x% of particles are updated every step

I wrote a similar snow app a couple of years ago in Logitech Modula-2 for DOS. I used an array to hold the flake data. If a flake stopped when it hit another flake, it would go again on the next round because the flake it hit would also keep going.