However, the main system is obviously needed. It has jobs to do. It still carries at least part of the game logic and its general subsystems and the graphics in whole must be communicated with to alter its states. At the very least, it must know alternations in camera orientation.
Now, what comes intuitively to me - and I'm not sure about it and it's part of why I'm asking - is that the main system shouldn't use the whole PCI-E bandwidth available at any given moment in the game (or simulation). It appears to me that it's using it mainly for intercommunication rather than row transfer of data (after the initial mass preloading of assets). It sure must be using only part of the bandwidth.
But I have an idea that probably answers it: The effect of performance by even a slight decrease of PCI-E lanes may be explained by considering that the game might be in need of bursts of transference of data. What I mean is, the game may need to be informed - or be in a state that this should be done - that a few textures must be switched around. Now, even if that process might be in "mass" of data quite thin, it might still benefit with a faster "burst" of data about that process. So, while the effect might be minimal - which is shown by all or most benchmarks - it might still be a benefit to have as much lanes - or bandwidth with a better bus version - as possible.
Am I right?