--reml-bivar-nocove flag changes result drastically

Hi,
I am calculating the genetic correlation of two traits using --reml-bivar.
One trait has about 1325 samples and the other has 2060 samples. The 1325 samples are almost all included in the 2060 samples.

I tried the default --reml-bivar and also the one with --reml-bivar-nocove. I found the rg were quite different between the two calculations:
#without --reml-bivar-nocove
rG 0.411151 0.302067
#with --reml-bivar-nocove
rG 0.133317 0.069778

As I don't fully understand --reml-bivar-nocove from the manual, I am wondering if it's ok to use it here, which gives better significance.
I paste the full results below. The heritability of each trait is not significant, but the values are consistent with or without --reml-bivar-nocove.

Thanks very much.
Yue


##
1) full result: without --reml-bivar-nocove
Source Variance SE
V(G)_tr1 0.459986 0.162264
V(G)_tr2 10.304661 4.956823
C(G)_tr12 0.895140 0.644744
V(e)_tr1 0.133474 0.158090
V(e)_tr2 17.627252 4.903815
C(e)_tr12 -0.618968 0.633102
Vp_tr1 0.593460 0.023413
Vp_tr2 27.931913 0.876409
V(G)/Vp_tr1 0.775092 0.267478
V(G)/Vp_tr2 0.368921 0.176114
rG 0.411151 0.302067
logL -4754.619
n 3385

2) full result: with --reml-bivar-nocove
Source Variance SE
V(G)_tr1 0.440506 0.161542
V(G)_tr2 9.667618 4.927088
C(G)_tr12 0.275120 0.115744
V(e)_tr1 0.152525 0.157651
V(e)_tr2 18.254509 4.883378
Vp_tr1 0.593032 0.023371
Vp_tr2 27.922127 0.875634
V(G)/Vp_tr1 0.742804 0.266987
V(G)/Vp_tr2 0.346235 0.175284
rG 0.133317 0.069778
logL -4755.076
n 3385