I agree that if you identify more stimuli for the experiment, you would get a more accurate fitting. And you need enough time points in your functional dataset for parameter estimation, but the results from 3dDeconvolve would tell you how good the fitting is.
There are basically two ways to run 3dDeconvolve depending on your purpose. If you want to check the hemodynamic response functions corresponding to each stimulus, you run the program with your stimulus time files. If you want to detech activation regions and/or make constrast tests, you run waver first to generate your regressors and then run 3dDeconvolve to get the fitting and statistics.
Cheers,
Gang