Thanks for this, I had been wondering if I should replace the faceplate on one of my lines because my old ADSL Nation filter failed so I moved the MK3 onto the line I use with telephones, leaving no filter on the second line at all. Seems it could be worth getting another MK3 after all even though I do not have a landline on the second line?
Possibly, though hard to say for sure. If you have a weak signal due to a long metallic path, but little common-mode noise, then the filter just adds more attenuation without significantly helping with regard to noise, making the overall SNR worse and reducing sync speed. On the other hand, if you have a strong signal but tons of common-mode noise, the additional attenuation is insignificant however the noise reduction is very beneficial, giving a higher SNR. The effect of noise bursts also need to be considered, as it could be that the filter reduces sync rate under normal conditions, but also reduces the number of CRCs that occur during noise bursts.
When the Mk3 SSFP was introduced, which differs from the Mk2 only by a higher inductance common-mode choke, a number of people getting nearly 80Mb/s on VDSL2 observed a decrease in sync speed when changing to the Mk3. This was due to the added attenuation of the higher frequency tones caused by the Mk3. Openreach seem to know this, as if you compare the VDSL Mk4 faceplate with the G.Fast faceplate you will note that the G.Fast variant has fewer turns of wire on the choke. In fact, the common mode choke in the G.Fast Mk4 variant looks very similar to that in the Mk2, while the one in the VDSL Mk4 variant looks the same as that in the Mk3.
Basically, I'd say there are too many variables to give an answer that is accurate for all situations. The only thing that can be done is to try the different faceplates and see what results you get. I only suggested it here in case Weaver ends up not being able to get the Zyxel modems with integral filter, so he could compensate somewhat by adding one externally with a SSFP.
Edit: Thinking about it, Weaver's lines are long and rural. The signal is therefore heavily attenuated, yet I can't imagine there are many noise sources along it. I wouldn't be surprised if Weaver's lines benefit from having no common-mode filter at all. It might be worth trying the VMG3925-B10B in this case, which has the good BCM63168 chipset but no filter (Johnson also has a custom firmware for it).