Hey Gavin,
(The following is all assuming the fee structure in NG functions correctly, which I'm doubtful of)
NG is a scaling solution in the sense that it allows the network to scale up further while getting away with it. Similar to the way weak blocks aswell as IBLTs accomplish this.
In NG, the information describing a full block / key block may propagate the network at constant time, reducing orphan rates to factors unrelating to the block size. The problem where scaling up by increasing the block size limit is thought to lead to miner centralization as a result of an increase in orphan rate, is solved by NG. Consequently, NG would allow the network to scale up without with no such miner centralization objection in mind; it is a scaling solution, at least in this sense.
As a rather trivial sidenote:
I believe it can be implemented as a (complex) soft fork. By placing microblocks in the next keyblock (enforced through consensus), and the next miner paying 60% of the fees back to the previous block's coinbase output(s) would make for a structure that is completely backward compatible. Including for SPV wallets. The microblock's ordering in the next keyblock is deterministic (as they reference their parent), which would allow a key-block announce to retain its constant-size, and constant-time propagation. A keyblock propagation message would include a standard 80 byte header message, along with a 32 byte last-microblock-generated header, which would allow any peer to deterministically reconstruct the full standard block, by plugging in the microblocks they should have already received.
Key-block / micro-block propagation may happen on top-layered p2p messages and enjoy the highly optimized NG protocol advantages, while still offering full backward compatibility to the existing infrastructure.
(There's some caveats still, but a soft fork is - I believe - perfectly possible)