When the Fluid Motion is off, we don’t apply any additional operations on the velocity field to remove divergence. However, with large voxel sizes, the influence of the source particle system on the driven particle system can have averaging effects that make it look more “fluidy”. Imagine a voxel with size of 1x1x1 generic units with 100 particles from the Realflow base sim falling in the same voxel. All their velocities will be dumped into the same voxel, and all particles from the driven particle system that fall into the same voxel will get the same velocity. As the size of the voxels goes down, you get a lot less particles into each voxel, and particles in those voxels end up following the source velocities much better. However, if the velocity magnitudes are very large, a driven particle could easily end up being kicked out into an empty voxel with no velocities at all. So particles would stutter, start and stop between frames.
For that reason, I consider the Stoke Particle Simulator a tool for creating new motion from existing motion, and not for following verbatim.
Btw, another way to fake partitioning with a PRT Loader is to just add a Noise modifier on it, change the Random Seed and save multiple times. I even had a Magma setup that could partition such a set up automatically by wiring the Seed of the Noise to a Magma input, but you can probably do it by hand, too. However, the PRT Cloner Repopulation likely produces smoother results.