Many moons ago, I invented* a rule that "you can always make people feel what you want about a #. either use percentages where they don't make sense, or whole numbers when a percentage does"
I hear it when I read 7 px -> 6 px means 14%(!!!!) less likely to find the horizontal/vertical only drag area.
Fitts's Law is logarithmic, not linear, and at these sizes the dominant factor is whether the target is discoverable at all, not its sub-millimeter width. "14%" smuggles in precision that doesn't exist in the underlying motor reality; it takes an imperceptible physical change and launders it through a ratio with a small denominator to produce a number that feels alarming. You could just as honestly say "we moved the edge by 0.097 mm**" and nobody would blink.
* I think? It feels like there'd be prior art on this
**
ppi = 262
inch = 1/ppi
mm = inch \* 25.4
# 1px ≈ 0.097 mm ≈ 0.004"
14% over estimates it because the user isn't clicking with uniform randomness, their clicks are normally distributed about the center of the line.