From individuals reading of the Eklund et al paper, would it be appropriate to drop SPM voxel-wise down to p<0.01 (consistent with the default p-value setting for FSL) and then just have p-FDR non-parametric cluster-wise <0.05?  Unless the clusters are quite large, I'm not really seeing how anyone is going to find anything with a modest sample size and default SPM p<0.001 and cluster non-parametric corrections.
 
Jeff