More test results for public. We ran my original query and found out
that on 4 cores CPU:
1 thread test 8.2 bits 8.3 by 10..15%
4 threads 8.2. wins 8.3 by ~2%
8 threads 8.3 finally wins 8.2 by ~2%
the same data set was affected during multi-threaded runs, but it's
100% cached from the disk. I guess we see the result of concurrent
scans optimization in 8.3...
--
Vlad