So an attacker could mine maximally expanded blocks completely filled with a single coinbase transaction with a huge number of outputs? Might this have an outsized impact on the network? (E.g. on the ability of nodes to keep up with verification)

They would sacrifice the block reward, although it may actually be a cheaper spam attack than via txs ArticMine

I'm imagining this technique being combined with the "big bang" attack and I wonder what the cost/damage calculation is with this

If you mine your own blocks, tx fess become irrelevant because you get them back anyway (assuming you don't broadcast your transactions)

It seems to me that having a consensus limit on coinbase outputs would inhibit such an attack and provide an upper limit on the number of miners which could mine on a p2pool instance. This would lead to several smallish p2pools, which might be good from a decentralization perspective

coinbase outputs are very fast to verify - they don't use RingCT

so there's no need to limit them

I guess maybe self-spend transactions are heavier than lots of coinbase outputs though

Ah ok

A miner trying to spam can just add normal txs with arbitrary fee

So it’s just being a miner that makes spamming more efficient

Are we talking here a big bang attack as an add on to 51%?

Not necessarily 51%, but just the fact that mining some portion of blocks could make spamming much cheaper in those blocks