I have been reluctant to respond to this question with hoping somebody else would say something.
Assume that you had added enough 0's at beginning to reflect the number (4?) of runs and that you had changed the number of 1's and -1's when you switched from max_lag of 4 to 3. I feel the results are very strange since such a small change of max_lag usually would not lead to a different contrast results. But, without knowing the experiment design and other details, it is hard to say anything about the situation.
Gang