When authoring workspaces that process large datasets, as Sampler can limit the data while testing - such as the first 10 rows (ignoring the rest). If you are utilizing feature caching - the transformer continues to process the remaining rows of data through the 'Not Sampled' port, which takes longer (and increased the cache size).
To improve performance of the Sampler, it would be handy to toggle the 'Not Sampled' port.
The PointCloudFilter has a similar ability to toggle Output Unfiltered Points.