This is why losing SatoshiDICE bets are so harmful to the system. You get back 0.00000001 or so on a losing bet. With such high transaction volumes, SatoshiDICE is handing out lots of difficult-to-spend dust spam.
I'd say it's the opposite of harmful. If one gambling website can harm Bitcoin, there is a problem with the system that needs addressed.
No-one is saying there isn't a problem with the system that doesn't need to be addressed.
We absolutely do need a way to somehow discourage the creation of small-value outputs, and the fact that we don't yet is a flaw in Bitcoin.
Actually, we do have such a system. It's the priority score system. Been there for some time, and the effects of Satoshi Dice would be far worse in it's absence.
One possible solution would be to simply make any transaction with an output value less than the minimum fee non-standard, so that those transactions won't be relayed or mined by miners. After all, if the value is less than the minimum fee, there isn't any rational reason to spend the output. (the fee is zero with sufficient priority mind you, but zero-fee tx's probably won't get mined forever)
We have this rule, also. Or rather one like it as part of the priority system.
The bigger problem is that it's only the size of blocks that is the limited resource - the incentive structure should make transactions that reduce the UTXO set cheaper, but actually coming up with rules that effectively do that without causing other problems is hard.
Actually, not hard, and being done. The many-outputs-transaction type is one such example, as it permits anyone that desires to pay a lot of different people at once to do it in a single transaction, which does reduce netowrk resource usage on a per-payee metric. This one does require a fee bsed upon it's total bytesize, but it's demontratablely lower than if one had to pay a minimum transaction fee for a set of transactions to pay to the same number of output addresses. Obviously, we won't really see this kind of thing being used much until there is a real market to get into blocks; but once we do sttart seeing them, they could be useful for weekly payrolls or monthly bill payment clients (like paying for your rent, heat, water, etc in one motion rather than a different transaction for each payee)
