That said, SelectCoins aught to be improved; ideally, it should never include an input if the fee for adding that input is greater than that input's value.
Or maybe it should just ignore inputs below the 'spam transaction' threshold for now, as being more trouble than they're worth.
https://bitcointalksearch.org/topic/m.1038709
I knew I saw that somewhere. That's an Intersango owner/dev I believe.
An end user's rational self-interest is total transaction fee minimization. I think this means that some approximation of total transaction fee minimization has to become the dominant coin selection method.
And the current coin selection solves some other problem. I think the way to look at it is this: you have some age-weighed inputs, some fee associated with a send from them, a discount rate that you are willing to trade a current transaction fee reduction for a future reduction, I suspect you might need estimates for average transaction size and time between transactions: the function to minimize would be the sum of discounted transaction fees, and in the absence of a clever solution, a stochastic/brute force/heuristic solution will do the job.
btw, I don't think a 'recent coin' premium can survive market forces. The miners who set the transaction fee, I assume will be constrained by block size, meaning that transaction size is the 'scarcity' that will determine fees. Which means that miners have an inventive to charge a linear fee per transaction size alone since coin age doesn't concern them (too much - I guess there is some risk premium for reversed spends).
For the vast majority of people this isn't really a problem.
Tons of tiny bitcoin transfers are a good thing for most people as they get to keep them.
We're fairly unique in that we dont keep any part of the transfer (not even a small fixed fee).
For most people though the incentives are to not spam people with small transactions.