One problem
may be that 3dDeconvolve's baseline is b0+b1*(n-nfirst). This means that the b1 (linear) term does
not have zero mean. You should account for this by using (b0+0.5*b1*(N-1)) for the denominator, rather than just b0. (Or you could use the "-polort 0" option to 3dDeconvolve to make the baseline model be a constant, but I don't recommend that!)
Also, if you actually want % change, you have to multiply your formula by 100, or it is fractional change. If you are doing this in 3dcalc, then I'd also recommend saving the data in floating point format, using the option "-datum float". This will avoid trying to scale the results to fit into a 16 bit short.
bob cox