Intraclass Correlation Coefficients: Significant Difference when H0~=0?

I'm trying to determine whether two intraclass correlation coefficients are statistically significantly different. The catch is that the two samples are between subjects, and the null hypothesis is that the difference between rho1-rho2 = .5. I've done a Fisher's Z' transform, but the confidence intervals are not consistent with the p-values (this does seem to work when the H0=0).

For example, rho1=.685 (n1=26), rho2=.382 (n2=54). The Z'1 I'm getting is .838, Z'2=.35. Where do I go from here?

Any information you have on how to compare these correlations is greatly appreciated!!