There are various ways of dealing with the problem of the very large number of voxels in your statistical image.
For rigorous statistical thresholding, you will one of these options:
Random Field Theory - see PrinciplesRandomFields.
False discovery rate - see PrinciplesFalseDiscoveryRate
Permutation testing - see PrinciplesPermutationTesting
Bayesian correction - see the articles listed under http://www.fil.ion.ucl.ac.uk/spm/doc/biblio/Keyword/BAYESIAN.html
Sometimes you may want to use a more liberal threshold that does not allow strong control of false positives - for example to report 'suggestive' rather than 'significant' results.
A perfectly reasonable option is to use a reduced threshold, still correcting for multiple comparisons.
Another is to report your UnthresholdedEffectMaps.
Another method that has been widely used is to use an UncorrectedThreshold. In my (MatthewBrett) view, this method can be extremely misleading to your readers, and should be avoided.