The difference between granted and claimed credits is not directly related to the speed of a machine. Several factors can influence it, e.g. the project, hyperthreading, Boinc version, etc... plus the same parameters for your wingman for projects not distributed in no-redundancy mode.

In fact, except for machines which are consistently claiming far too low or far too high, some WUs are granted more than claimed while some others are granted less, for the same device.

You might have a look at these two threads for discussions about some real life situations (but they do not cover all possibilities)
Claimed vs Granted
Claimed vs Gramted Credit

Cheers. Jean.

Ah, higher than, lower than, so are you saying that the actual grant per hour for the faster machine is lower than the slower? Any 64 bit OS/Client here involved? Both 32 bit?

Both CPU's run at maximum Ghz? How about you post both client's Benchmarks and a couple of jobs, full quorum detail of same project from each machine.

On the Result Status page click on the work unit name. That will open a second window with a list of all the copies of a particular work unit. That allows comparison to what what the wingman/wingmen returned.

Same project, yes, several HCMD2 job e.g. done by each of your devices. Then we see run times, claims and grants and can reconstruct how them also in relation to the present benchmark values, Whetstone and Dhrystone (aka Flops and MIPS), mark each so we know which or devices did the jobs.

A program such as CPU-Z will tell you the real Ghz a processor is running at during actual crunching. Some processor tend to throttle down when only doing crunching since they are identified as 'idle/nice' processes, but don't think it's the case here.

You want to get to the bottom, but as per Jean's post, that may simply not possible. Could be these different CPU brands may in addition not have the same performance on the different projects, cycle by cycle.