We want to know which clusters are significant (multiple comparison corrected) at p < 0.05
So according to the alphasim result above, should we threshold the result maps in AFNI at a) 0.025 or b) 0.05 ( ie: 0.025 X 2 - to convert 2 tailed to 1 tailed)

First of all, you must be running an older version of 3dClustSim -- the program was changed in Feb 2015 to allow for the differences between 1-sided and 2-sided testing. Also, some bugs were fixed this summer. So you need to update your AFNI binaries -- for example @update.afni.binaries -d -- if you have privileges to write into the AFNI binaries directory!

When you do this update and re-run 3dClustSim, you will get multiple tables. The -help output explains them at length. Briefly, you get all 3 NN cases, and for each NN case, you get 3 "sidedness" cases: 1-sided, 2-sided, and bi-sided -- the latter means that positive and negative voxels above threshold are clustered separately but both sets of clusters are kept (unlike the 1-sided case).

The 1-sided table is used with 1-sided thresholding (e.g., positive only t-stats, or F stats) and the 2-sided table is used with 2-sided threshold (e.g., t-stats with pos and neg kept together). You have to decide which table is appropriate for your analysis. If you are loading the tables into the activation statistics dataset, and then using the Clusterize plugin, then the plugin will decide which table to use, based on your settings in the GUI.

Many people don't know that the right-click popup menu from the top of the threshold slider lets you set positive-only and negative-only thresholding (the default is 2-sided). And there are other options on that menu as well that you might find useful.

Finally, the choice of where to set the per-voxel p threshold is also yours. Then 3dClustSim tells you what the cluster-size threshold should be for a cluster-level significance (alpha).