When there are multiple random-effect terms in a model, the simple concept of the R^2 breaks down. That is, there is no natural or unambiguous concept of the usual R^2 with two or more variance-covariance terms in a model. Various authors have proposed R^2-TYPE statistics, based on either the marginal or conditional residuals, or based on differences of log-likelihoods. I suggest you read Liu et al. 2008. J. Appl. Stat. 35:1081-1092, or Vonesh et al. 1996. Biometrics 52:572-587. One has to calculate these brute-force with SAS, by first outputting the residuals into a data file.