Do we want to keep the maximum block size large enough that all transactions can be included in every block, or do we want to keep it small to increase tx fees and incentivize mining?
Good question.In my humble opinion, the block size should not be arbitrarily limited as it is now (1MB is the limit; typical blocks these days are 30-250K big), but should 'float' -- miners should collectively decide how large a block they're willing to validate and build on top of.
Obviously a miner wants to include as many fee-paying transactions in their blocks as possible, until the fee paid is less than their cost of validating and including the transaction (which is a small cost).
But miners also don't want to spend a very long time validating other miners' blocks, so they have an incentive to ignore blocks that are outrageously big. If they were willing to build on a 10-gigabyte-big block that took ten minutes to download and signature check, then they're shooting themselves in the foot-- an evil miner could mine a huge block, and then get a head start on mining the next block while the rest of the network was busy validating it.
BUT: moving to a floating maximum block size determined by miners will be really hard; it will require everybody-- merchants and miners and users-- to upgrade. It may never happen, because other ways of supporting very high transaction volumes might develop before then.