Setting a cluster parameter in a sharded cluster, and immediately executing a command on the same mongos that set the parameter, may result in the command seeing an old value (or default). 

This is because the mongos writes the cluster parameter to the config server, but does not refresh its cache correspondingly. This refresh only happens on startup, when executing getClusterParameter, or periodically (default 30s) due to the 
ClusterServerParameterRefresher.

It seems reasonable to expect the mongos which updated the cluster parameter to respect the new value in sequential operations.