This is a common question. The 99.93% reduction seems unrealistic, but it is not. Translating from physical modeling into words and pictures, the statistic is based on an average room volume of 150 cubic meters with the crosswind unit running at 1 m/s, for a 5-minute close interaction of 0.5 meter between the worker and resident.
Some basic arithmetic will help in this explanation.
To simplify, it is a matter of comparing the most infectious condition against the least infectious condition during the interaction. The most infectious condition occurs around the first minute of interaction in the left room [view animation here in Part 3] where the diameter of the expanding cloud is approximately 0.6 m. Therefore the volume of that cloud is 0.1 m^3. So the ratio of the cloud volume to the room volume is 1/1500. That is where the '1500' number comes from in Question Q8, and creates the 99.93% statistic (0.9993 = 1 - 1/1500). Given that you can see in the right room how the cloud is completely dispersed by the crosswind, the initial infectious cloud density is transformed into a complete homogeneous dispersion into the room volume without ever first touching the resident in its original high-density form, the heart of this statistic. It is the equivalent of keeping worker and resident distanced at a great distance of far more than the 2-meter rule, while in reality being only a half meter from his resident. And this is even before the filter stack takes effect to scrub the dispersed particles (see Question Q8 for how the filters work).
Anytime a worker violates protocol of maintaining the air stream between him and his resident during a close interaction, that 99.93% is degraded. That is why good adherence to the protocol is important to maintain that improvement.
Now here is something even more interesting related to your question: the closer the worker is to the resident, the better is the statistic. It feels counter-intuitive. However, consider this: if the distance is 0.3 m instead of 0.5 m and the unit is running at 2 m/s or more making sure that the dispersion condition in the right room is maintained, then the statistic becomes 99.98% improvement! The reason is that at the closer distance without a crosswind unit, the probability of infection would be greater by about 2X. However, in both cases, the aerosol is completely cleared away from the resident.