The Ruckig trajectory smoothing plugin extends the duration of all waypoints if smoothing isn't successful. It's an iterative process until success is achieved. Attempt smoothing --> extend the duration --> attempt smoothing again.

The code is here:

https://github.com/ros-planning/moveit2/blob/cc635471aadfb9446398ece319ae31c6b72bec86/moveit_core/trajectory_processing/src/ruckig_traj_smoothing.cpp#L357

        for (size_t time_stretch_idx = 1; time_stretch_idx < num_waypoints; ++time_stretch_idx)
        {

The resulting trajectory could be faster if only the failing waypoint durations were extended. There's no reason to extend all of them at every iteration.