Just a brief conclusion of this series of discussion. So for whole-brain analysis,
(1). For voxel-wise methods, either FWE or FDR, with p < 0.05, is definitely ok.
(2). For cluster extent threshold, a primary threshold (i.e., CDT) of 0.01 is definitely not ok, 0.005 is also risky, and only 0.001 or more stringent value is suggested. Then FWE, FDR, or Monte Carlo simulation in 3dClustSim is all ok to calculate the k value.
Hope some experts can confirm these conclusions.