Created after a discussion on the mailing list : https://groups.google.com/forum/?fromgroups=#!topic/mongodb-user/qkiQmLDRuBA

For each incoming connection, mongoSs create a connection to each shard primary (and one to a secondary is using appropriate flag).

This is really blocking for scalability since each primary will have as much as all incoming connection to the cluster.

Usage of these connection really internal stuff and should (and could easily) be pooled.
Of course it reveals also that a primary is required for any kind of request.