afni.nimh.nih.gov

I think these results have to do with using the -nptr option. Half of your trials begin at the beginning of the TR, and half begin midway throught the TR. Each of the two types of trials contributes alternating points in the IRF. Since h0, h2, and h4 are based on some trials, and h1, h3, and h5 on other trials, you might expect their norm. std. dev. to be more correlated among timepoints derived from the same trial.