Sharding Pitfalls Part II: Running a Sharded Cluster` highlights common mistakes when running a sharded cluster at scale, including waiting too long to add a new shard due to high utilization, under-provisioning config servers which can lead to performance issues and impact all queries in the cluster. Another pitfall is using the `count()` command on sharded collections, which can produce inaccurate results due to optimizations that bypass key pieces of the sharding functionality. To mitigate these pitfalls, strategies such as manual balancing of targeted "hot" chunks, adding a shard at low traffic times, and disabling balancer temporarily are suggested. Additionally, ensuring config servers are load tested and slightly over-provisioned is crucial for optimal performance.