I'm thinking about writing a minimal-dependency C library implementation of Bitcoin. There's the other thread here where people are reverse engineering the protocol for a python client. More and more alternative implementations will appear when Bitcoin gains more momentum, especially since the stock implementation can't possibly cover all the Bitcoin use cases (think of a laundromat that accepts Bitcoins for example). Alternative implementations will happen and decent protocol standardization is the only way to prevent potential problems with multiple implementations inter-operating on the network. Multiple implementations will make the network grow faster, standardization will make the network more robust. Both will dramatically increase security, as a lot more potential problems will be discovered.