This problem with N-body has been around for a long, long time - as documented in the thread I have cited here:
https://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=4295#67559

And it's amazing to me that the project has virtually ignored all the reporting on it. But your post regarding 100% CPU usage has convinced me to allow N-body tasks again to test it out. Hopefully, it's a game changer.

A preventative workaround fix is to set the CPU time usage to 100% in the computing preferences, if set lower than that. After doing this none of the n-body tasks that I have run have stalled on the PC's that previously did stall, which all have Intel i7 CPU's.

As noted in my earlier post If an n-body task has already stalled the only fix I know is to exit BOINC and restart.

Note setting CPU time usage to the maximum 100% will likely increase the CPU temperature. I was able to bring it down by eliminating overclocking and setting performance options to maximize stability and save energy.

I changed to 100% CPU time usage after seeing the above post and haven't had the problem since doing so - probably around 100 N-Body units completed flawlessly. Was having the problem on 30% to 50% of N-Body units previously.

Well done corysmath! Now, if we can only get the N-Body developers to read this and think about a program fix. (But I'm not holding my breath.)