I think the subject says it, but I think it's awkward to have to type:
cov2cor(data.matrix(data)) - cov2cor(model$objective@info$expCov)
It seems like it would be simple to have a class (is that the term?) for the residuals function that operates on an mxobject. I often find myself looking at the residual covariance matrix to see where misfit lies.