The current implementation will first total the number of particles in all the designated partitions by inspecting their headers. Then, it will load the desired percentage of particles starting at the first partition in the list until the correct percentage has loaded.
Would it be more useful to have the N% evenly distributed over all the partitions?