In sharded clusters, you'll find that writing to a mongos becomes the bottleneck on the system. In order to make this faster, it would be nice to be able to parallelize it by allowing these tools to write to multiple mongos's and distribute the load across them.

This could work in the form of reading in a seedlist of mongos's to write to. The tricky thing here is how to handle cases where one of these mongos's or an operation fail.