Yeah, but who wants to run that server and guarantee perfect uptime and results for a year? Who wants to verify that whoever ran that server didn't do anything funky? A smart contract is a much better option all around.

As to the suitability of the platform, granted, it would essentially be a test net for tracking real money, but I'm not suggesting Eos at full capacity with user-provided contracts or anything. I'm talking about a chain with just a few contracts built in, rigorously tested, demonstrating how the user-provided contracts will work and interact in the final release.

The hard parts in implementing Eos, the parts that will take a long time, are the novel parts: getting the VM ready for untrusted code, getting the parallelism working stably, optimizing all of that at the end...  all the stuff no one's ever done before. The components necessary for the token sale, on the other hand, are comparatively small and not particularly novel, and thus will be faster to implement.

(Edit) Of course, these are just predictions and opinions, and I could always be mistaken. But I'm still thinking there's something here until someone convinces me otherwise.