Please read the help section for 3dttest++ titled
HOW COVARIATES WORK -- that explains exactly what is going on.
If you use the default centering option '-center DIFF' for a 2 sample test, then
#0 SetA-SetB_mean = difference of means [covariates removed] will just truly be the difference of sample means (at each voxel). However, different options for centering will alter the "mean" results, since the "mean" values are then calculated by linear regression of the data samples w.r.t. the covariate(s) and then projected to the chosen covariate center.
If you are using covariates AND are using a 2 sample test, you should read the section
***** CENTERING (this subject is very important -- read and think!) ******* because it in fact is very important.