My understanding:
Input Dataset = your data
Stim timing = a 1D file where a 1 represents the onset of where you want to start averaging. So if you're doing event related, you pretty much just leave your stim file alone, but if you're doing block, you should change the file so there is a 1 at the beginning of each epoch. (Like its one big event).
then set maxlength to the number of time points after the onset that you want to average.
So if you want to average the 2 time points before the onset and 8 time points after, you'd change your timing 1D file so that there is a 1 two frames before the onset of your stimulus. Then you'd put max length of 10.
But I am just guessing from the relatively brief help file.
probably best to do as dr. cox suggested and use 3dTcat / 3dMean
I have a couple scripts that could be useful, if you're interested.