The Oracle machine and code runs 3 times faster than IBM's alternative and that leapfrogging happened in just four months? Wow. How would you like to be the IBM sales guy trying to sell against Oracle right now? The only guy envying the IBM salesman right now is the guy from HP.
IMHO TPC-C is like a Formula 1 car race. Both have absolutely nothing to do with "real life". The TPC-C description states: "Each warehouse tries to maintain stock for the 100,000 items in the Company's catalog and fill orders from that stock." In the "real world" no mentally sane CIO would accept such a massive hardware for such a tiny problem.
Sorry, but today for 100.000 items I do not need disk space or database at all. Even 100.000 items of 10MByte each -- HiRes photos supplied -- is just 1TB. Even a single CPU slot standard PC server board supports 128 GByte of ECC RAM. So, put 16 of these in a cluster and run the whole shop completely in memory.
3X performance in 4 months is simply not possible. They basically did it with a much larger and more expensive machine. This per system/cluster metric is very questionable. Is a supercomputer one system? That surely will beat all records.
Using the $/performance metric, former record (IBM) is $1.38, Oracle is $1.01, 27% better. However, IBM system was available 10/13/2010, but the Oracle/Sun system won't be available until June 2011, i.e., 8 months later. In 8 months, system performance will probably advance more than 27%. So Larry's big announcement really is not that big a deal.
Indeed there are many ways to manipulate benchmarks and I am sure the big three know them all. The good news to me inm this story is that Oracle continues to back Sparc, Solaris and Java, keeping those techs viable.