I think bittorrent would be a really good model to follow.
But it's not the only path to success; Perl is a good example of a successful technology with One True implementation and no specification beyond the One True implementation.
Then again, development of Perl 6 seems to be going really slowly.
And I think breaking up the functionality is a really good idea. For example, if the algorithm for signing transactions and the format for public and private keys were standardized it would be possible to create a Bitcoin iPhone app that stored the private keys (wallet) on the phone and submitted signed transactions to a proxy that was connected to the p2p network.
I see the functionality broken out into these pieces:
1. p2p nodes that are constantly connected and relay blocks and transactions
2. transaction monitoring ("tell me when there are new transactions that match some set of criteria")
3. new block monitoring ("tell me when there are new blocks that match some set of criteria")
4. transaction validation (answers question "is this transaction valid, and how many validations does it have?")
5. block validation ("is this block valid, and how many validations does it have?")
6. bitcoin mining: race to generate a new block and earn ฿
7. wallet storage (generate new addresses, and store their public/private keys and any transactions that correspond to them)
8. generate new transactions (sign ฿ with private key(s) and submit to p2p network)