It seems that 32-bit version of hp9 is still considerably slower than 64-bit version, at least on win7 x86 and x64 respectively. The difference is like 0.26cpd vs 0.38cpd on i3-540@3.6.
mikaelh, are you going to do something about it? Would be great.
There's not much that can be done about it. I think it comes down to modular exponentiation being faster on x64.
Also as I have said before, haveged is probably not useful for mining. I don't see why the mining process would need entropy. It shouldn't really be detrimental either except for the fact that a few CPU cycles are being wasted.