Chris,
So you want to see whether the coupling at a few regions under one condition is different under another condition? If the trials vary between 8.5-10 seconds, it's probably not going to make much different the duration difference is less than a TR, but dmBLOCK with -stim_times_AM1 would not hurt either.
I don't have a better suggestion, but maybe you could try:
1) run the typical regression analysis with 3dDeconvolve with option -cbucket, then use the following to isolate those effects you want to remove from the signal,
3dSynthesize -cbucket CbucketFileFrom3dDeconvolve+orig -matrix x1DFrom3dDeconvolve -select ... prefix EffectsOfNoInterest
and finally obtain the signal corresponding to the specific condition:
3dcalc -a InputFor3Deconvolve -b EffectsOfNoInterest -expr 'a-b' -prefix CleanData
This is not ideal because, as I mentioned before, the trial-to-trial variability can't fully modeled. Plus the assumed HDR function is also a crude approximation.
2) PPI is slightly different from what you're looking for (and it's also a very crude approach), but it may be worth trying?
Gang