Tx fee prevents another type of attack (namely spamming the memory pool) or at least raises the cost. However there is no required tx fee so if there was no maximum block someone with sufficient mining power for example could make a 1TB block (all free txs made by the miner) and now the blockchain is >1TB.
It's extremely likely that a 1 TB block would be orphaned well before the miner transmitted to his first peer.
It is just an example. 1TB was hyperbolic but multiple smaller (but still large) blocks over time could still accomplish the same goal. A well connected attacker could analyze orphan and latency and choose the optimal size which produces the max blockchain growth over time. An attacker with less than 51% of hashpower but looking to degrade Bitcoin wouldn't necessarily care about the cost. An attacker with 1% of the hashrate would still have >500 attempts per year. A larger attacker could optimize the attack by not broadcasting until he was at least 2 blocks ahead in a modified form of the selfish miner attack but with a goal to bloat the chain instead of gain increased rewards.
The 1MB limit will almost certainly be raised but using another sanity cap is a good idea. Optimally it would be some floating cap which is deterministic and based on blockchain usage but that may take some time to develop and test. In the interim raising to say 10MB gives the network breathing room while limiting the scope of an attack.