This may be true in general, but for the particular hosting company where I run my VPS, disk is the limiting factor in the decision when I have to upgrade to a more expensive plan. With pruning, you can not fully support the network and you have to be careful with your wallet handling, as rescanning is not possible. That's actually what I do on one of the VPS exactly because the blockchain is already too big for the plan I use there, but I would have preferred to run an unpruned full node if I could. Increasing the block size makes this issue larger by orders of magnitude.
Furthermore, as stated above already, block relay optimisations also help only "little" with bandwidth. (At best a factor of two in comparison to a much bigger increase in block size being discussed.)
Yes. Your situation re disk is different, and the pruning has known improvements to come.
Block relay optimization is more than "just" a factor of two gain, because unconfirmed tx are relatively uniform over a 10 minute window (so the network has an impressive capacity to handle these, as seen by how fast the mempool can grow), whereas blocks need to be sent at once, so ~600x data / second. Also, up-link is invariably slower than down-link which is a big concern for miners publishing blocks.