The intent of 'well-partitioned' is to minimize writing by more than one node to the log device (i.e. keep writes to one instance). You are still going to get shadow locks and cache page movement as you are in a cluster.
You should be using the MDA tables for monitoring - tables such as monCIPC can help here. Combined with other MDA tables, you can identify the SPIDs and connections in use. Remember, the SPID is cluster-wide and is a modulo number based on the owning instance - e.g. in a 3 node cluster SPIDs 1,4,7,... belong to instance 1...
What is your current 'low' value for CIPC and large CIPC buffers and why can you not increase to the suggested value? The impact will be slower performance most likely due to higher waits on resources.
Are you giving enough resources to the cluster?
Chris