EDIT: Well actually the fee does present a legitimate problem, and it would be hard to make this system work without any incentive like that. Hmm...
The incentive would be that even though it does not bring income, it reduces cost (in the sense of making the unspent outputs smaller).
A stronger counter argument could be that colored coins might get harder to track, on the other hand this stuff would work probably better on a merge-mined chain anyways...
One of the other questions (which is still under debate) is, if block space should be limited - if it is limited, "defragging" the unspent transactions is going to be costly and either has to be enforced by rules (e.g. "after latest 1 year with no transaction there HAS to be a refresh, to allow for a sliding blochchain") or it will happen out of necessity (having more unspent TX is more expensive than accepting regular transactions) of miners. If block size is unlimited (or practically unlimited), something like a shorter sliding block chain is easier to implement, as the next limiting factor then is probably rather the bandwidth (and refresh-transactions can be precalculated, so other nodes only need the TX-hash) between miners and network than block space.
The edge case would be to have these transaction in every block, meaning every block would contain a full set of all addresses with a balance - downloading all headers since genesis and 1 block is enough. However, one either needs then to precalculate a lot of refreshes (as a big part of blocks will be refreshes) and just receive the tiny delta of actual transactions each new block or a really fast internet connection.
The longer the chain becomes, the more additional "junk" is stored (transactions that could be pruned) but the smaller blocks can get due to fewer refreshes. The current chain is the other edge case - all blocks are needed to get the current set of unspent outputs and 0 refreshes are in the chain.
