A measure to reduce the frequency of the hangs is to use little swap space.

Here's an extract of my system log:
syslog.7.gz:Sep 28 19:20:00 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.7.gz:Sep 28 22:19:03 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.7.gz:Sep 29 04:47:32 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.7.gz:Sep 29 07:16:44 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.6.gz:Sep 29 13:35:04 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.6.gz:Sep 29 13:35:04 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.6.gz:Sep 29 17:27:41 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.6.gz:Sep 29 21:43:55 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.6.gz:Sep 29 21:43:55 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.6.gz:Sep 30 03:30:53 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.6.gz:Sep 30 07:50:10 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.5.gz:Sep 30 12:48:52 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.5.gz:Sep 30 12:48:55 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.5.gz:Sep 30 22:07:09 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.5.gz:Oct  1 04:49:01 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.4.gz:Oct  1 11:07:24 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.4.gz:Oct  1 13:33:58 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.3.gz:Oct  2 10:59:56 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.3.gz:Oct  2 13:52:15 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.3.gz:Oct  2 13:52:16 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.3.gz:Oct  2 21:44:36 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.3.gz:Oct  3 01:08:25 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.3.gz:Oct  3 09:26:54 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.3.gz:Oct  3 13:30:52 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.3.gz:Oct  3 17:25:08 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.
syslog.3.gz:Oct  3 21:44:55 quimper rtkit-daemon[2231]: The canary thread is apparently starving. Taking action.

Up until Oct 3, I got 2 to 6 hangs per day (each lasting minutes or hours). In these times I was using between 6 and 10 GB of swap space (in addition to the 16 GB of RAM my machine has).

In the evening of Oct 3, I killed and restarted a process which was eating several GB of virtual memory (plasmashell - a well-known memory hog). Now, the last 3 days, I'm using less than 1 GB of swap space - and zero hangs happened.

So this leaves two hypotheses:
a) The hangs occur when a lot of swap space is in use.
b) plasmashell not only leaks memory but also some other kind of resources. When these resources get scarce, the hangs occur.