The TPC-C IO load was seriously getting out of hand to the point of distorting the server system architecture required to run the benchmark. Do you remember the original Intel Merced platform that did not see the light of day? It had 24 PCI slots. This is unavoidable in a write (IUD) intensive benchmark, evening by increasing the tpm-C to warehouse ratio. Writes must be written to disk eventually.
TPC-E significantly reduced IO load probably by increasing locality of hot data, rather than data size and by shifting to a much more read intensive query set, which can be cached to eliminate IO.
I am not inclined to think either is realistic, as 200 plus disks on a 2 x Xeon 5460 is bigger than anything I have seen. I guess real apps tend to have highly CPU intensive queries, so may be 100 disks on a 2 socket might be better