I haven't advertised the -clustsim option in 3dttest++ because I'm not finished testing it yet. However, it
does give bigger cluster thresholds than -ACF directly. My belief is that this effect is due to the non-stationary smoothness of the noise, coupled with outliers in the data. With this method, I get pretty accurate false alarm rate (FAR) control when using the Eklund et al. comparison vs. the Beijing FCon-1000 datasets. Using ACF, the FAR still comes in too high.
But, as I started with, I'm not done with this subject yet -- but also, I'm so busy with other things I'm not sure when I'll be able to make further progress. I'm just back from one trip overseas, and leaving on another in a week. Plus just getting over the damn flu.