Lee,
With your current design, you should not have any problem of multicollearity when you run 3dDeconvolve with the convolution results although there is some overlapping among the HRF's. However, this does not say anything about the sensitivity of the corresponding linear system.
Regarding the randomization of experiment design, see HowTo3 on AFNI site: [
afni.nimh.nih.gov]
Gang