Liz,
Thanks for the clarification.
In the model, the beta's depend on two things: the original signal and the regressors. Depending on how you pre-process and normalize the signal and also depending on how you create the regressors, the absolute values of beta can be arbitrary. However, the relative magnitude of the beta's should be roughly the same regardless of the pre-processing and regressors. This is why people usually convert their beta's into % change and then do group analysis or make comparisons.
When you calculate the % change of beta's (don't confuse it with the % signal change you did during the pre-processing), you don't need scale them with your regressor magnitude (i.e., 1.604) since your irf's have a peak of one.
Hope this helps.
Gang