I could see it both ways. Requiring each device to overcome noise on its own makes having multiple data jacks (not just for stacking NR, but to direct connect to more devices) and implanted commlinks more useful. I'd still apply NR to both ends of a conversation, so a DR 3 cyberdeck running signal scrub could cut through 5 points of noise, and then the DR 3 device on the other end of the connection would also benefit from the signal scrub and be able to communicate through 5 points of noise as well. It also makes modding a non-master device to add Vectored Signal Filter a viable tactic for more than just unattended devices.
On the flip side, it adds a bunch of minutiae to figuring out what is affected, and the noise system is already muddled enough. The receiver dongle for commlinks specifically states that it affects slaved devices as well, while VCF does not, but VCF (being a cyberdeck module) can be added to any device, where they have made it very clear in comments on the forums that the intent is that commlink dongles only work on commlinks (and not RCCs or Cyberdecks, much less generic devices). So its possible that they worded it like they did since the VCF might be used in a device that could not become master of a PAN, or it may be that the VCF simply does not provide its NR to other devices in a PAN if it is attached to a device acting as master.
For simplicity's sake, I'd say that the strongest device wins, whether part of a PAN or not. In real world logic, if you have really good signal filtering software, you can pick out a signal better, compensating for a less capable transmitter, and when receiving, the higher DR device may simply have a higher gain antenna to cut through noise and thus be able to transmit to lower DR devices.
'Noise" in game is an abstract covering a number of things from distance to poor signal to overloaded bandwidth and even intentionally disruptive traffic (DOS). There might be some cases where it makes little sense that a higher DR device can compensate for a lower DR device, but I'd imagine that in the majority of cases it could. So if you treat Noise as a conglomeration of varying effects and choose to apply one simple rule rather than myriad exceptions, it argues in favor of a high DR PAN master being able to stave off its DR in Noise for all of the slaves.
This is, notably, different than the effect of Noise on matrix actions where it is applied as a negative dice pool that is only mitigated by NR and not DR. So a DR 6 'deck subjected to Noise 6 would still function (and be able to interact with all of its slaves), but all Matrix actions requiring a roll would be subject to a dice pool penalty of 6-NR. Defense and resistance rolls are excluded per the rules on page 231. Furthermore, the Noise level is only considered at the roller's location. So if you are hacking a device in a high noise area, but you (or, more accurately, the antenna of your 'deck) are not in a high noise area, then you would not suffer the higher penalty. And vice versa. If both you and the device are in the same noise area, it is not applied twice, so I am pretty sure that this (this paragraph) is RAI.
As for any interpretation making cyberware useless, keep in mind that when Noise exceeds DR, it simply disables wireless functionality. Working through my cyberware, only my Smartlink would be completely useless (I think). My datajack would still work, just not wirelessly. Cyberlimbs don't benefit from wireless, and most cyberware retains its core function without wireless. I'm also assuming that part of the installation of cyberware includes a direct connection so that even without a datajack or trode net you could still operate it (even wirelessly). Or have I completely misread the gist of your statement (in which case, whoa, what a tangent I just went on)?
