Refreshing Alternatives

Given how important the concept of scarcity is in defining a game based around the mechanics of resource auctions, you'd think that there'd be an obviously correct way of implementing it, but I have yet to find One Way to Rule Them All.

Right now, I'm looking over the available alternatives, and I'm coming up with different ways to model scarcity. Let's list the pros and cons for each:

1. Direct Expiration: in this model, which is the default one from the Quick-Start guide, each trait is simply unavailable for bidding after being bid for a set amount of time. Pros: simple, requires no randomizer, correctly models scarcity by making things unavailable when they would like to use them; Cons: lots of numbers to keep track of, players may not take "dramatic" liscence without breaking the rules.
2. Approximate Expiration: in this model, we replace the set timer based expiration discussed above with a probabilistic timer. For this model, at the end of each round, the player rolls a d100 for each trait that is not currently available for auction, and if the roll is less than some number (determined for the column, so there's only like 4 numbers to keep track of), then that trait is refreshed. Pros: only 4 numbers to track, very simple, accurately models scarcity; Cons: lots of potentially tedious die rolling (O(n)--it grows linearly with the number of traits spent, and is done every round), introduces dice into an otherwise diceless system.
3. Exact Buyback: in this model, we do away with the whole expiration model, and we replace it with a point-based refresh system. After you spend a trait, it does not naturally refresh, unless you pay a certain number of points for it. These "points" are abstract units only used for the buyback system (this may involve dice). You get a certain number of them each round, and you can spend or save them as needed. The cost is associated with the column, so there are only 4 numbers to track. Pros: simple, no repetitive die rolling; Cons: introduces new mechanics that are not used elsewhere, may allow for the RPG version of button-mashing because it does not properly assign a higher cost to more desirable traits.
4. Inexact Buyback: in this model, as the Exact Buyback model above, you have a cost and all that, but you don't get to choose which trait comes back when you pay the normal rate, one at random from the column refreshes (based on a die roll of some sort), but they can also buy back a particular trait at a higher cost. Pros: can be managed entirely by players, penalizes button-mashing; Cons: requires strange dN die rolls, still has dice involved.
   
5. Probabilistic Blind-Auction Buyback: in this model, instead of having a set price for a particular trait, we have a Uniform Random Variable in a range indicative of the relative difficulty of refreshing (basically, you have a die-roll and the refresh rate, and some formula that turns the two into a number). What the player does is, as he gets more points back after his round is over, he has the option of buying back any traits he has listed as spent. However, the way that he does this is by bidding a number of his points on each of them, and for each trait, you roll a die, and come up with a cost. If his bid is less than the cost, he loses those points and gets nothing, if it is equal or more than that cost, he gets the trait back. Pros: great way to model cost--with more valuable traits costing more, can be done by players without the ST doing the rolls (provided that they don't cheat on the rolls); Cons: requires 1 roll per trait, may just annoy players.