Pnoise analysis with windowed sampling

Dear experts,

To improve my understanding of LTPV analyses, I was simulating a simple track and hold circuit shown below. I am using spectre 23.1 ISR11 for simulating the design with PSS analysis performed with shooting Newton method. Capacitor is 1pF and clk signal goes from 0 to 1.2V.

For case (a), I am using a uniform sampler with sampling frequency of 1MHz. Pnoise is performed with noise type time average as well as sampled. Output frequency is swept from 1 to 100MHz and maxacfreq is set to 5GHz. Simulated total integrated noise over 1 - 100MHz in timeaverage case matches closely with sampled noise integrated over 1 -500kHz and this is inline with my expectation (about 56uVrms).

I repeat the same with a windowed sampler, where the sampling clock is generated by gating 1MHz clock with a 100kHz control signal (case (b)). Result is sqrt(10) increase in the noise floor.

How do I interpret this increase in the noise floor both qualitatively and quantitatively?