Hi George,
You might try a deconvolution analysis in which you have 3 stimfiles corresponding to the onset of set size 1 trials, set size 3 trials and set size 5 trials. Estimate each IRF using a minlag of 0 and a maxlag of however many TRs it takes to get 10 seconds past the probe. You'll get an IRF, for many areas, which has two peaks, one corresponding to the encoding period and one to the probe period. You could then compare the area under each of the two peaks, or compare various intervals within the IRF. In our experience, although the sternberg effect of response latency is observed in the probe period, differences in activation by set size are predominantly found during the encoding period.
Sally