My understanding is that it tries to evenly split aff/neg for large teams, the idea being to prevent, say, a school with 8 teams having all 8 be neg at once, with associated impacts on the coaching staff. It would be most useful in a division with a large percentage from a single school, or at least with several schools that were much larger than average. I'm not sure exactly what computation Tabroom does to consider a school "large," but I suspect it's computed as a percentage of the field, so "larger than average" is probably more meaningful than an absolute number of entries.
The downside is that if Tabroom is attempting to do this kind of balancing (in presets or non-side-locked powermatched rounds, it won't, say, break a bracket just to balance aff/neg split), it undermines the "randomness" of the pairing.
Whether to use it comes down to how nice you feel like being to coaches of large programs, and whether you think it matters that all the entries from a school are on the same side or not, vs. how much you care whether the pairing algorithm is closer to randomized.