Exactly. An attacker won't ever have a split on their malicious chain because there would be no latency between their nodes. As a result, the more splits the main network has, the less percent of total hashing power an attacker would need. If we assume the worst case that block propagation time will eventually be as slow as BGP (which takes about 1 minute to propagate worldwide), the network would only lose at most 10% (1 min/10 min) of hashing power by chain splits.
Further to this, do nodes have any individual incentive to propagate new blocks? Either carrot (i.e. they get coins for it, are prioritized to receive new block updates or some other incentive) or stick (i.e. they're seen as bad neighbors if they don't and their neighbor nodes drop their connections)?
Nope. But if they do that, they might as well just not listen for connections like most of the network is doing, as it would have a similar effect. If it became a problem, though, something could be done about it.