You normally need to make your FIR window cover the period of likely non-zero HRF, so typically somewhere between 20 and 30 seconds.  However, with your stimulus spacing you will have regular overlap that I doubt you'll be able to separate (normally timing is jittered if you want to look at the peristimulus changes in detail).  So you probably are better off just using the standard FEAT peristimulus outputs, as for this case it is probably as good as it is likely to get.  If you do want to continue with the FIR analysis, then I would probably use a 12 second window and make each FIR function of width 2 seconds, since there isn't anymore information that you are likely to be able to extract given your TR and the non-jittered timing.  The result of the analysis will then be a separate parameter estimate (PE) for each FIR EV, where each one represents the value at one particular peristimulus offset.  To see the whole peristimulus plot you need to combine them together, and you can do this with fslmerge and then look at the "timeseries" in FSLView to see the peristimulus output at each voxel.

All the best,

Mark