> Does the percent of 'blocks of no interest' compared to total blocks affect the analysis?
If you have some decent number of blocks of interest (e.g., more than 5), you should be fine.
> Can I concatenate all the runs across sessions and do the analysis
> instead of doing the analysis on each run separately.
As shown here (http://afni.nimh.nih.gov/sscc/gangc/CD-CorrAna.html), those preprocessing steps should be done for each run separately, while the final regression analysis can be performed on the concatenated data (and regressors).
> Is there something I should be careful about when I concatenate all runs across sessions.
Follow the steps on the aforementioned webpage.
Gang