Sorry for not getting back sooner.
The problem with sign randomization on just 6 datasets is that there aren't that many sign combinations. With a total of 16, I can add permutation to the mix, which would make the calculations combinatorially possible. However, I'm literally working on that code today, and have to test it, and then package it up for mass consumption (e.g., write some help for it).
The alternative that I see for you, if you want to try this out now, is to output the residuals from 3dttest++ with the -resid option, then run 3dFWHMx with the -ACF option, use the 3 ACF parameters in 3dClustSim -ACF, and then use the cluster size thresholds that come out of that method.