You have a very high-level idea of what you want, how big you want it to be, and what it will look like. You do not, however, have usability requirements, cost/server requirements, specific security requirements, and so on. Therefore you throw out a vague suggestion about your potential project, with assumed technologies that you've heard about from friends/blogs/etc, and plan on simply explaining the idea in your head to a programmer, hoping they can build it.

One of the commonest things to do with savings is to lend them to some Government. In view of the fact that the bulk of the public expenditure of most civilized Governments consists in payment for past wars or preparation for future wars, the man who lends his money to a Government is in the same position as the bad men in Shakespeare who hire murderers. The net result of the man's economical habits is to increase the armed forces of the State to which he lends his savings. Obviously it would be better if he spent the money, even if he spent it in drink or gambling.

The man who invests his savings in a concern that goes bankrupt is therefore injuring others as well as himself. If he spent his money, say, in giving parties for his friends, they (we may hope) would get pleasure, and so would all those upon whom he spent money, such as the butcher, the baker, and the bootlegger. But if he spends it (let us say) upon laying down rails for surface card in some place where surface cars turn out not to be wanted, he has diverted a mass of labor into channels where it gives pleasure to no one. Nevertheless, when he becomes poor through failure of his investment he will be regarded as a victim of undeserved misfortune, whereas the gay spendthrift, who has spent his money philanthropically, will be despised as a fool and a frivolous person.

That said, I think it can get difficult to get several dialects of the same language in. For example English, the is currently only one language code and that is "en".

1 and 2 is doable using the current system:
Upon receiving some funds you can create an temporary address/private key pair, send the funds to this new address, and use the generated private key to generate the two outgoing transactions. Now store both transactions and throw away the temporary private key.
That way the original address does not vanish and so no coins will be lost.

The project would be strictly limited to improving bitcoin-related open source projects with an initial primary focus on improving the usability and security of the bitcoin client and a secondary focus on other improvements (such as bugfixes) to the bitcoin client.

The biggest issue for me is the security of holding other people's money. If anyone knows a better way than transferring coins to offline encrypted wallets with emergency key backups printed on hidden sheets of paper, I'd love to hear it. Perhaps I can find three trusted people and give them each 2/3 of the key as an emergency backup in case I get hit by a bus.