...
In a nutshell: use for VBM OR add TIV as covariate, never for fMRI, usually never for quantitative MRI
...
In a nutshell: never for fMRI, maybe for repeated measures VBM and quantitative MRI
In some designs (multiple groups, 2 sample or ANOVA), you can specify "ANCOVA-by-factor" regressors. The factor would usually be the group, and the covariate the global value (mean over the mask).
"These options allow different subjects to have different relationships between local and global measurements." If not repeated measures, it will do the group x global effects.
References
Dartmouth wiki: useful compilation about global scaling.
...