I have done a lot of performance analysis of the ESS and what is good for
different types of data.  In your case I would have recommended at least 4
LUNs but not more than 8 placed each on different arrays and use simple
RAID-0 striping across this set.  Make sure half of the LUNs are on Cluster
1 arrays and half on Cluster 2 arrays.  I call this methodology 2
dimensional striping and have done a lot of work with the IBM performance
lab on this.  What happens is you have now spread the IO across 4 to 8 times
as many disks.  The performance implications can be dramatic.  However, if
the workload cannot generate IO fast enough the ESS begins to think it is
RANDOM and little benefit is gained.

I expect the storage pool to be a predominately sequential process.  Thus,
you can expect some very significant improvements.  Though these
improvements will only scale up if you are using fibre channel.

Keep in mind though I have not profiled the IO of TSM to a storage pool, but
generally speaking you want to spread your data across both clusters to take
advantage of the write cache and resources of both clusters.