Be very careful comparing the plain numbers! You have different error signal slopes so the corresponding frequency noise level for the sensing might be totally different !
Anyway it is good to compare individual parts which should be equal, e.g. the RFPDs at RF frequencies (or mixed down using the same(!) LO power (or better: mixing/conversion gain)
So what you say in the first sentence is right, but don't forget the gain of the amplifier! You can't compare those without taking that into account.
The increase from 30nV/rt Hz to 1uV/rt Hz is a bit higher as i would expect it but still makes sense. The minimum gain for the ZFL-500LN amplifier is 24dB, the difference between those numbers is ~30dB.
We should measure the mixer output with the amplifier but without the photodiode (input terminated) to see where the noise floor of the amplifier is. Then you know how far that is below the RFPD noise.(and can calculate the RFPD noise level from that i\f you want)
You say that the noise at the mixer output is the same for both loops, but the setup is completely different (different mixer, different LO power (23dBm and 7dBm or so). So be careful. It shouldn't make a big difference but you have to measure the RF noise around 35.5MHz or use the same mixer setup, e.g. plug in the ACAV RFPD in the RCAV mixer and compare then. Then you have real good numbers for comparison of the PDs.