Thank you for your notes on the matter.
1. They still have to run the equipment to perform the mining. This has a cost associated with it, and they need enough revenue to keep it going. The more revenue their competitors get, the more hashing power their competitors can afford which increases the profitability of their competitors and increases the mining difficulty. Eventually the attacking miner (or pool) is operating at a loss. When they lose enough money, they can't afford to continue to run and have to shut down.
Mining is already extremely centralized and there is evidence that one company could already control over 50% of the hashrate directly or indirectly. There are profits to be made in raising the blocksize from this principle miner or a cartel of miners who wish to shutdown competitors elsewhere while they can further centralize mining in a single location like this where they aren't concerned as much with orphaning blocks due to increased latency:
http://www.datacenterdynamics.com/content-tracks/power-cooling/bitmain-to-build-a-large-cryptocurrency-mine-in-xinjiang/97265.fullarticle2. They lose out on the transaction fees from the transactions they refused to include. This is lost revenue, and other miners or pools that choose to include those transactions will be able to afford more hash power, thereby making the attacking pool less relevant.
This miner, Cough...Bitmain... cough, would still be extremely profitable because most of the block would be filled with others tx fees and they would simply top it off occasionally when they wanted increase the blocksize. THis would actually likely increase their profit margins as the txs sitting in the mempool would than compete in a fee market which would drive up tx fees.
3. If their block gets orphaned, then all those "spam" transactions can be confirmed by some other miner or pool. As such, they can pay a very hefty fee for playing such a game and taking such a risk. Eventually this additional revenue for the "honest" miners and pools will allow them to afford more hash power, thereby making the attacking pool less relevant.
I would disagree , they would be encouraged to continue down the path we are already witnessing- centralizing mining further to reduce orphan risk by placing most mining under one roof -
https://bitcoinmagazine.com/articles/bitmain-reveals-plans-for-major-bitcoin-mining-data-center-in-northwestern-china-1478797051/ , collecting rent fees from other miners who wish to be hosted in this facility, pretending that mining is becoming more decentralized by creating contracts with "other" companies , advocating for blocksize increases to onboard more txs and collect more fees while preventing competition from remaining profitable.
The only way miners could be successful in such an "attack" would be if there was a cartel of miners (or pools) that controlled significantly more than 50% of the hash power. They could then all agree to exclude low-fee paying transactions and avoid mining each other's "spam". If this occurs, then bitcoin is already a failure.
Miners order txs and users validate them. Miners work for economic users. The current Miners are ultimately powerless to the will of these users and bitcoin will carry on with or without them.