Don't bother -- your site can't even exist, Kano has decreed that the whole thing wasn't possible.
Seriously though, if the thief is willing to go through enough trouble, he can launder the coins, but
that suggests a certain amount of understanding and sophistication on the part of the thief, and it's
quite possible they won't bother, which makes the tracking efforts worth a try IMO.
I agree that if a thief is willing to go through the trouble, they could launder bitcoins, so as to make them very hard to track; possibly even impractically hard to track.
I don't have a strong opinion on whether 'marking' coins - basically, trying to keep some record of which 'coins' were stolen (or rather, which fraction of the balance at a particular address was 'stolen') would have the effect of reducing the incentive to steal bitcoins.
I do like the idea that even if a small fraction of bitcoin users will refuse to accept 'marked' bitcoins, then this instantly decreases the value of 'marked' bitcoins; and this can possibly have a knock on effect; so that 'marking' doesn't have to be total, or centralised, for it to have an impact.
But I also see the argument that it would be very difficult to know who to trust to maintain lists of marked bitcoins; there seems to be some centralization inherent in the idea of lists of bad coins.
And there would be conflicts of interest; the more bitcoins that get marked as 'stolen', the more valuable non-stolen bitcoins become; anyone with a lot of bitcoins would be incentivised to have other bitcoins 'marked'.
There is also the fundamental difficulty of establishing whether bitcoins that are alleged to be stolen, were actually stolen, or not. I'm not talking about any specific case here.
If its possible to pay for goods, with Bitcoins, and then later declare the bitcoins used in the payment to be stolen, and hence marked, you mess with the way bitcoin handles non-repudiation.
It would also totally change the setup of services, which have user accounts layered on top of the bitcoin protocol - they don't have a direct mapping from individual users, to bitcoin addresses - while they can be considered to be outside the bitcoin system, in some sense, 'marking' would not work well with them, in practice.
Like all these economic things, it's very hard to reason about the effects of such a system, so I've no strong opinion on it.
The fact that the complete history of a balance is stored, and publicly available, allows you to think about doing interesting things like this, in a way thats hard in other setups.
It might be interesting, if, instead of balances, there were specific 'coins' in the protocol (at the moment, balances lose their individual identities, when they pass through a transaction) - that would allow 'marking' to be done properly - maybe such a system could support 'marking' in some decentralised fashion, and be more resistant to theft. I don't know, but interesting to think about.