The Reported Load Averages experiments and tests of these web pages are based on the assumption that (on my servers, at least) the background loads (from cron jobs and such) are close enough to 0 to ignore.
This web page provides the details for what the backgrouns load "really" are. ("really" in quotes because it is not clear what is to be trusted.)
First, for the kernel with the commit, from the load Vs number of processes test:
The test had completed, but the reported load averages were recorded for another couple of days, with the server not used. Zoom in on that data:
That the 15 minute reported load average never goes below 0.05 is a finite number of bits math issue. Otherwise they seem to spike up about 0.1 every so often.
A historgram of the data:
Now, for the control kernel CONFIG_NO_HZ=n. First the overview, including the load Vs number of processes test:
As with the previous test, the reported load averages were recorded for a couple of days after the test completed. Zoom in on that data:
The differences between the two are interesting, perhaps worth further invesigation. Anyway, ignoring the background stuff adds at most about +0.1 -0.0 uncertainty to the reported load averages exeriments.