Mariko,
Since I don't know much about your experiment, I can't say which (the continuous 6 minutes or the interspersed 3 minutes of fixation) would be a better choice for modeling the baseline. Just a reminder: In AFNI you don't need to specify the time points for baseline modeling because they are implicitly indicated by all the regressors. In other words no need to create a separate regressor for baseline modeling.
You might use the full F values and the -fitts option in 3dDeconvolve to check which choice of baseline modeling gives you "better" results.
Gang