The problem is the network cost is tiny, yet has nothing to do with the long-term cost of storing the UTXO set, and also is fixed so that profitability for larger, more centralized, pools is always higher than smaller pools.
There is a marginal cost to the miner for increasing the UTXO set in the form of capital investment of memory and fast storage to store it in. When the UTXO set gets large enough to be a problem miners will have an economic incentive to reduce their hardware costs by favouring transactions that shrink the set over those that grow the set.
Even the miners with lower capital costs will have an incentive to limit the size of the set because it affects the speed at which other nodes can validate their blocks and thus their orphan rate.