Prime95 uses half of each core?

[Mini-Geek]

I recently got a dual-core, and I noticed something. I have two Prime95's running with the -An switch, and if I stop one of them, the other (according to taskmgr) uses half of each core. I always assumed it would have used one core to 100% and the other to 0%. If it really does use half of each core, is it losing efficiency by doing so instead of using one core 100% and the other 0%? What's keeping it from using both cores fully instead of both halfway?
Are the services not setting affinities correctly on boot? I ran the two Prime95's with the -An switches, then let them create services. I can see the two services listed (as Prime95 Service-0 and -1), but they each seem to just point to Prime95.exe with no -An switch. It has created the correct work0000.ini, work0001.ini, etc. files in the Prime95 folder.
If I set the affinities in the Advanced menu (still on the windows that the services opened up, btw) and stop the one that should be running on the second core, taskmgr shows ~25% usage on core 0 and ~75% usage on core 1. If I stop the one that should run on core 0, it has core 0 at ~0% and core 1 at 100%, as expected.

Or is taskmgr just reporting it wrong when only one of the processes is running?
Also, I noticed that the two cores get approximately the same CPU time, but I would've expected that the second core wouldn't get used as much (because of programs, especially games like BFME2, that don't support dual-core), and the second process would get more CPU time.

I suppose this really isn't a problem, but it's strange behavior, and I was wondering why it's like this.

[retina]

It is just how the task scheduler works in windows. At each time slice interrupt the OS choses which core will continue on with the task.

As for efficiency it will usually mean that the L2 cache requires purging before switching that task to another core. The OS might purge L2 anyway (whether switching cores or not) when the time slice is finished so perhaps the efficiency loss will be constant regardless of which core is used, but there is a chance that on some processor models that keeping a task on the same core can permit the OS to leave L2 untouched. Nowadays with shared L2 this may make no difference. I guess only a test will show if the OS behaviour is smart enough to know when to purge L2 and when to leave it to maximise efficiency.