I'm referring to these images:
from 1718's beta testing:
http://www.fightingpi.org/Resources/...12_Day_9.shtml
Basically, the Victor 884 curved away from being linear in one direction, while the 888 seems to curve a little bit away in the other (thus over corrected when adjusting the 888 to be more linear than the 884).
The result would be the exact opposite of the 884 - you would be able to finely tune the speed of your motor at the low end, as each step change in input would be a smaller change in output, but would have worse control at the high end, as the same step size would result in a much larger change in output.
Of note, the data you gathered didn't seem to show this same result, which leads to a question of which data set more accurately reflects real-world usage on a FIRST robot. Only having two data points makes it tough to reach a conclusion on this.