I don't have your mask here, but a back-of-the-envelope calculation with the following script,

3dClustSim -nxyz 39 39 40 -dxyz 3 3 3 -fwhm 6

shows the following table:

# CLUSTER SIZE THRESHOLD(pthr,alpha) in Voxels
# -NN 1 | alpha = Prob(Cluster >= given size)
# pthr | 0.100 0.050 0.020 0.010
# ------ | ------ ------ ------ ------
0.020000 55.7 62.8 71.4 78.7
0.010000 35.1 39.6 45.2 50.0
0.005000 24.5 27.4 31.8 35.3
0.002000 16.6 18.7 21.5 23.6
0.001000 12.8 14.6 16.9 18.7
0.000500 10.1 11.6 13.5 14.9
0.000200 7.6 8.8 10.4 11.7
0.000100 6.0 7.2 8.7 9.8

So it looks like you need a minimum cluster size of 28 voxels to survive the threshold of cluster level significance of 0.05.

However, since you blurred the original data by 6mm FWHM, the real spatial correlation would be most likely greater than 6mm. With a wild guess of 8mm, you may need a minimum cluster size of 51 voxels:

3dClustSim -nxyz 39 39 40 -dxyz 3 3 3 -fwhm 8

# CLUSTER SIZE THRESHOLD(pthr,alpha) in Voxels
# -NN 1 | alpha = Prob(Cluster >= given size)
# pthr | 0.100 0.050 0.020 0.010
# ------ | ------ ------ ------ ------
0.020000 103.4 116.1 133.6 146.5
0.010000 64.2 72.9 84.2 91.8
0.005000 44.0 50.2 58.9 64.5
0.002000 29.2 33.4 39.1 43.6
0.001000 22.0 25.7 30.1 33.8
0.000500 17.3 20.2 24.0 26.5
0.000200 12.8 15.1 18.1 20.3
0.000100 10.2 12.3 15.1 16.6

Gang