Your post is rather vague. Are you reporting an observation or just guessing?
There are several reality factors that modifies the simplified I^2 torque relation. Winding resistance however doesn't affect it. I conclude from your post that you don't yet understand the physical background of the equation and neither its limitations.
In geared drive, gear friction will be a major source of torque variation, I'd expect that it's also temperature dependent to some extent.
Stator saturation occurs at higher currents will probably show a certain temperature depedency.
I guess that due to gear friction, it's hard to get better torque accuracy than 10 or 20 %.