wbmw,
Doesn't this just prove my point that SPECint is not a good indicator for databases performance?
No. You just need to apply the gray matter, and it does not take a lot. A simple constrained of a test with the servers equipped with the equal amount of memory would reduce the discrepancy. Further, not crossing the flat addressable memory limit of one architecture of just one architecture, and not other would reduce the discrepancy.
Your benchmark (TPC-C) is not a CPU test, it is a client server system test, with so many unconstrained variables beyond the CPU that it can't be used to prove much without putting some constrains on some variables.
Joe
Market Data 
Markets 
AI
