Thanks robbolito for the tests; you have always been kind to providing games whenever requested so personally, I dont need to see them -- just want to make some observations.
Elo difference amongst the different systems between Rybka 4.1 and Houdini 2.0:
AMD Phenom(tm) II X6 1090T: 60
Intel(R) Core(TM) i5 Q- 750: 88
Intel(R) Core(TM)2 Quad Q9550: 133
It is normal for one program to do better on certain architectures, but difference for the Q9550 is a bit high even given error margin, and considering it is being speculated the true Elo difference is ~64.
For 736 games, error bar is +/- 18, so 100 difference is normal.
Edit:
Looking at
http://www.clemens-keck.de/base.htm, Houdini 2.0 is ranked 3023 (+/- 11), and Rybka 4.1, 2949 (+/- 9). If we factor in the errors, including the one for 736 games, you get 130 Elo difference. Now, I guess it doesn't look so abnormal after-all.