The reason this is so sensitive is that you're effectively "erasing" the exposure, along with undeveloped fog.
To oversimplify, when silver halide grains are exposed to light, they develop "latent image specks" -- nanoscale specks of metallic silver, directly reduced by light exposure. These specks catalyze the reduction effect of developing agents, so that only halide grains with latent image specks on them will develop.
Your pre-development ferricyanide bath is bleaching (some of) these latent image specks back to effectively unexposed halide, and/or directly dissolving them (if you use a bleach that includes fixer, like Farmer's Reducer). Because these latent image specks are so tiny, there's a very fine line between "improved contrast and shadow value separation" and "no image remaining". There's both very little silver present, and a rather high ratio of reactive area to mass in each individual speck -- hence the small window between "does nothing" and "leaves nothing" in which you're getting an effect you want.
Generally, you'll gain more with less risk of undoing all your dodging, burning, and contrast filtering by developing and fixing, then bleaching as needed to clean up highlights or, as you note, improve shadow values separation.