The current games have performance bottlenecks and even single-point problems, high-performance clusters (HPC) that can not scale linear scale-out and highly available clusters (HACs) that are highly consistent (anti-crack) and IDC-rich. Therefore strictly speaking can not be regarded as a true distributed cluster computing.
To solve problems with online: special game servers are developed with open source code and a new compression protocol.
For ontable games, this is not critical. And in 3D games where you need a ping max of 5 ms, there are several solutions, 1e was invented even in the first network games, when the objects were calculated by the prediction method, rather than a complete miscalculation of physics.
Yes, all low-level coding developers, anderstands how we are all limited to the architecture of the x86 and the binary of the calculus, probably it was someone's joke when developing the first computer to use 0 and 1, rather than 0..63 for example, and in our time we are optimize the code and find new solutions, limited to x86 architecture, this applies to all manufactured silicon processors.
Yes, we learned to transfer some of the calculations to video cards, but this is not enough.
Linear scaling also does not solve the problem, I agree with you. New solutions are needed.