WilhelmGGW,
If your Ubuntu is an Ubuntu 64 and your XPs XP32 this is normal and I mentioned it several times in other threads. I have both running alternatively in the same Q6600 quad. As expected the flops benchmark is the same in both cases but the integer one is 50 % higher in the 64-bit mode than in the 32-bit mode.

And probably the HCC WUs have much more integer computation than WUs of other projects. In the above quad a HCC WU needs an average 3.75 hours under XP32 and little less than 3.0 hours under Ubuntu 64.

If your machines are all running in the same mode, either 64 or 32 then that could be really Linux which makes the difference, unless it is the processors' architectures?

Cheers. Jean.

Aside from how much faster my Ubuntu machine is on pretty much everything, my particular observation with respect to HCC is this:

When BOINC downloads a work unit, it gives an estimate of how much run time it will take to process it. On my web page of validated results, I have a "final statement" of how long the unit actually took. (Or at least I think that's what I'm getting there.)

I've compared these two numbers for lots of work units on all my computers. My Linux machine is the fastest I have. That is, it processes all units significantly faster than similar units on my other computers, and the final time for each work unit is significantly shorter than the initial estimate.

But the real standout is in processing HCC work units. With them, the actual run time is far and away faster than with other projects. That's why I asked this question.

Follow-up question:
Does the WCG-host interface pick up on performance anomalies like the one I've reported here, so that it feeds each computer installation more of the project work units that run on it most efficiently?

That is, would your analysis pick up on my Ubuntu host performing extraordinarily on HCC, and send it more of units than it would, say, to my XP machines?