Andrew,

There are a few approaches to assessing the multicollinearity among explanatory variables in a model. One of them is the condition number of the design matrix that is used in 3dDeconvolve to alert the user about its severity. Some people do view that a correlation higher than 0.4 between two regressors indicate a potential multicollinearity issue, but that's not necessarily always the case. See more discussion about this and a couple of other multicollinearity assessment methods here:

[en.wikipedia.org]

Gang Chen
[afni.nimh.nih.gov]