FEAT offers the option of intensity normalisation (of all volumes in each time series in order to give constant mean volume intensity over time); however, this option is turned off by default, as it is considered that this is an over-simplistic approach to a complicated problem (see, for example, [7]).
We investigated the effect (on inter-session variance) of turning intensity normalisation on. It was found that this pre-processing step does reduce the overall fixed and random effects variance (on average by about 10%), and therefore slightly increases the fixed and random effects Z values (again, giving, on average, approximately a 10% increase in the number of supra-threshold voxels).