Dear Akiva,
Before SCP, we had been thinking hard about how to do sharding efficiently without degrading any security guarantee. A simple solution which splits the coins, or TXs in to several partitions will just not work. You have to answer more questions to have a good solutions. For example, I wonder in your proposal, if a transaction spends a "coin" that ends in "1" and creates a new coin that ends in "1", which partition should process the transaction? What is the prior data needed to validate that kind of TXs?
The problem with other proposals, and probably yours as well, that we see is that the amount of data that you need to broadcast immediately to the network increases linearly with the number of TXs that the network can process. Thus, sharding does not bring any advantage than simply using other techniques to publish more blocks in one epoch (like Bitcoin-NG, Ghost). The whole point of using sharding/ partition is to localize the bandwidth used, and only broadcast only a minimal data to the network.
Clearly we are able to localize the bandwidth used with our SCP protocol. The cost is that now recipients need to themselves verify whether a transaction is double spending. However, we think that it is a reasonable tradeoff, given the potential scalability that SCP can provides.