AFAIK, CIPC regular message pool buffers are used primarily to transfer the shadow locks between instances. The large buffers are used for the transfer of cached pages between instances.
If you upgraded from an earlier SP than SP130, make sure you read the 'What's New' and release notes:
http://help.sap.com/Download/Multimedia/zip-asecluster/ase_whats_new.pdf
Cluster IPCs use the private interconnect, if available, so the faster the better. In your case, you also might be using a lot of locks. Ensure that your application is 'well partitioned' and that you are not accessing the same tables from multiple nodes for R/W.
As of SP130, you can also look at SIDB, which sets up the database in the cluster as more of an HA environment. This means that the database can only be accessed from one node (so failover takes slightly longer as no locks get shadowed at all for that database). SIDB is more of an HA environment - but the application may perform faster. You will have to know your SLA's to use this, however, if you cannot take the downtime for the database to be recovered on the failoever node.
To also minimize the shadow locks, look into the granularity of the table locks. Not all tables need to be DRL. Some can perhaps be DPL (Do you have any tables still as APL? - make them DPL).
If you are performing batch updates on tables, you can also look into the LOCK TABLE command before performing table level updates (e.g. UPDATE...WHERE..), especially if the table is DRL locked. This can vastly improve the performance almost to SIDB or standalone, as only the 1 lock will be shadowed instead of millions - but concurrency could suffer on that table.
Chris