LT1366: LTspice noise simulation yield strange values

[This is a re-post. I deleted the first one due to a profile error.]

I am trying to compute the output noise of for a Butterworth filter built using Sallen-Key stages. I reduced the circuit to the following voltage follower to investigate incomprehensible noise results.

When:

R3D = 1.02Meg noise is 83300 nVrms
R3D = 1.05Meg noise is 1.67 nVrms (a significant decrease for a 3% increase in resistance)

So, noise decreases with increased resistance? That's the opposite of what I expected. The noise gain has not changed, so op-amp voltage noise contribution should not change;  op-amp current noise should cause higher noise with increasing resistance values; resistor thermal noise should also result in more output noise with a higher R3D.

I attached the .asc file.

6507.Butterworth-8th-Opt-Voltage - D stage noise.asc

Does anybody see what I am missing? The simulator results make no sense to me. Thanks.