See the sheer block size that would need to be implemented if you do the maths, and you see the magnitude of the problem.
i agree with some of the things you say but i am not yet convinced about either one of the solutions that are being proposed.
things are never as easy as just increase the block size and be done with it. increase by how much? the 8 MB of BCH? if we assume your numbers are correct that is 8*7=56 tx/sec that is still nowhere near 2000.
so what will we gain with that kind of increase? we just made running a full node 8 times harder.
i say on chain scaling is needed but i ask myself by how much? we can not have 500 MB blocks which is nearly 2000 tx/sec
that is where Lightning Network comes in. although that is less centralized than on-chain scaling but it is a solution to get that much tx/sec.