The build I have for your head uses a gold plated SMA launcher. Mine uses a modified Nickel/brass BNC launcher. As you can note from the pic the SMA launcher requires a SMA coupler to connect to the minicircuits SMA attenuator whereas mine requires an BNC-SMA adapter to achieve the same.
Given that the SMA coupler I used is from Kirkby Microwave as part of the cal kit for the VNA, I doubt it could be at fault as it has been quality checked before shipping.
That just leaves the SMA launcher itself as compared to the BNC launcher, the SMA launcher is not likely to be worse than the BNC launcher which is on my head. SMA carries a much higher frequency spec. than BNC. The 0.2dB of loss at 800Mhz looks like some kind of resonance dip as it improves afterward. That implies some parasitics at work.
There is also the issue of the 10dB cross talk reduction gained by using different schottky detectors in the new head. It is possible, that at that frequency, the cross talk between the monoblock combined ref. detector and the actual detector disturbs the leveling. But that's just a guess.
All in all though, I am satisfied that the design I have is more than adequate and that both designs do the job.
I think rather than concern ourselves about 0.1dB improvement here or there, the improvements in the DIY nature of the latest design matters more.
Given the availability of the components used in the new design as well as the simple build construction any decent home workshop can economically turn one of these new heads out and I figure that adds longevity to the Tek SG504 instrument lifetime and that's what the forum is all about.