To add onto Daniel and Zhark's comments. It looks like you are trying to estimate the smoothness of the data from the actual ANOVA results. This is... not ideal. Instead you should use 3dFWHMx to estimate the smoothness of your individual subjects errts files, calculate the average of this smoothness value, and use those values to put into your 3dClustSim command. Relevant warning from 3dFWHMx below:
*** Do NOT use 3dFWHMx on the statistical results (e.g., '-bucket') from ***
*** 3dDeconvolve or 3dREMLfit!!! The function of 3dFWHMx is to estimate ***
*** the smoothness of the time series NOISE, not of the statistics. This ***
*** proscription is especially true if you plan to use 3dClustSim next!! ***
