Ellen,
I think i know what is going on. When you want to do cluster analyses on a statistical map you need to include thresholding information aswell as clustering information. For your 3dmerge command:
3dmerge -clust1 1.42 1992 -prefix myfile MannWhitneytest
you are not specifying a thresholding value for your statistic. In this case all the voxels in the brain are 'significant' so when it tries to cluster, each voxel is connected to every other voxel giving you one huge cluster which is the size of an entire brain. This makes sense given the size of the one cluster that 3dclust found (756545) is enormous. This would also make it so that changes in the minimum cluster size around 1992 would have no effect on the map. Obviously when you are viewing your test within afni you set your thresholding slider to a specific p level. Get the Mann Whitney statistic associated with the p value you want, 2.0 for instance, and use that as a threshold when clustering.
Your new command line should be:
3dmerge -clust1 1.42 1992 -1thresh 2.0 -prefix myfile MannWhitneytest+tlrc
You may have to specify a subbrick in your Mann Whitney test to make sure you are accessing the correct one (see -1dindex and -1tindex in the 3dmerge manual).
I hope this is the answer to your problem,
Christine Smith