Hi Edward,
I assume that the script you provided is for just one run. You could concatenate all your runs into one file and run 3dDeconvolve on the concateneated file.
I don't understand why you want to flatten the baseline or normalize it. As long as you add "-peak 1" when you run waver and normalize your dataset before you run 3dDeconvolve, those regressor coefficients would reflect their corresponding % signal change. Indeed you could add "-polort 2" in 3dDeconvolve if you feel that a quadratic trend would be a better fitting for the baseline.
Gang