Why no correlation in OPUS solutions between X/Y/Z and Lat/Long error estimates?

I've run a number of things through OPUS now and I'm puzzling at the seeming lack of correlation between OPUS's error estimates for the X/Y/Z solution and the Lat/Long solution. I can see why they wouldn't be exactly the same, but they appear way off. It seems to me they should be at least in similar ballparks.

As an example, here are the OPUS error estimates for something I just ran (it's a backyard test next to the side of the house and trees so relatively high errors):

X: 0.062 m
Y: 0.089 m
Z: 0.067 m

Lat: 0.010 m
Long: 0.029 m
EL HGT: 0.126 m

I do see that their RMS values work out to be 0.073 and 0.075, respectively, so in that respect they're close.

If there's a simple explanation that I'm not seeing or pointers to an explanation, I'd greatly appreciate hearing it. I have the sense I'm missing something profoundly fundamental (i.e., stupid).

Tom