Sarasir,
There is no single return path in my mind, and I doubt that the BNC sleeves themselves are at the center of any particular current flow path. But there are currents flowing through the Z-plate and all the ground planes connected to it. These elements are in theory a common path, but to individual current loops they are just part of the roadway that comprises the loop. For DC and low frequency loops I view these elements as part of a wider highway with a fairly even traffic distribution that accommodates the needs of multiple loops. However, at higher frequencies, because of impedance variations due to interconnections and/or structural characteristics, the traffic distribution is no longer even. As a result the electron traffic will be more intense in the areas offering a lower impedance path, and of course less intense in those areas with a higher impedance. Nevertheless, every part of this highway will be affected by this traffic to one degree or another. The BNC sleeves are connected to it, so they are part of the highway, although as I said, I doubt that they are at the center of any particular path.
Of course, most of the traffic is comprised of differential currents used to power the various scope circuits. But some of this traffic, particularly HF traffic, are common mode currents that originated with the switchers due to stray capacitances, inductances, or radiation. Although unintended, these are also current loops, and are also what we call GND noise. Had these currents been confined within the boundaries of the switchers that originated them, by proper design, circuit layout, and/or shielding (like rf-loop says, don't cause noise), they wouldn't be an issue. As it stands, this didn't happen originally, and although it has been alleviated on current production, it hasn't been eliminated. There is probably little incentive to do so since competing products suffer from similar ailments of various degrees.
Now, however, I think that what you are doing can potentially confine the fast rising currents associated with each switcher to their respective boundaries. Basically, it seems to me that this is a major redesign. With the switchers isolated any common mode currents attempting to escape via the associated input or output wires can be effectively filtered with a common mode choke. Along with some additional differential mode filtering, I think these switchers can potentially be as clean as traditional linear circuits. With that accomplished the HF common mode currents on the Z-plate and ground planes should be virtually gone. At that point, in my view, the only impediment to a totally GND noise free scope will be the TFT. I'm not sure that a Faraday cage that doesn't include the screen can eradicate the effect of this noise source. However, from what I've experienced, this noise source is mostly near field, and usually can be easily avoided by keeping the probe cables away from it. So if it's not too bothersome, why mess wit it!