After having experienced this a couple of more times, i have found that it is not a 100% reproducible problem - more like 50%, and that is in my case.
Some more pieces of log that looks somewhat relevant (everything else says kernel):
nov 29 17:50:11 Xperia10III ofonod[3338]: Server: < AT+BCC\r
nov 29 17:50:11 Xperia10III ofonod[3338]: Server: > \r\nOK\r\n
nov 29 17:50:11 Xperia10III kernel: sec_ts 1-0048: [sec_input] [R] tID:0 mc:16 tc:2 lx:869 ly:173 v:0100 cal:A2 id(0,0) p:0 noise:0 lp:(0)
nov 29 17:50:11 Xperia10III dbus-daemon[2112]: [system] Rejected send message, 2 matched rules; type="method_return", sender=":1.68" (uid=100000 pid=5881 comm="/usr/bin/pulseaudio --daemonize=no -n --file=/etc/" label="u:r:kernel:s0") interface="(unset)" member="(unset)" error name="(unset)" requested_reply="0" destination=":1.13" (uid=0 pid=3338 comm="/usr/sbin/ofonod -n --nobacktrace --noplugin=,he91" label="u:r:kernel:s0")
nov 29 17:50:11 Xperia10III dbus-daemon[2112]: dbus-daemon[2112]: [system] Rejected send message, 2 matched rules; type="method_return", sender=":1.68" (uid=100000 pid=5881 comm="/usr/bin/pulseaudio --daemonize=no -n --file=/etc/" label="u:r:kernel:s0") interface="(unset)" member="(unset)" error name="(unset)" requested_reply="0" destination=":1.13" (uid=0 pid=3338 comm="/usr/sbin/ofonod -n --nobacktrace --noplugin=,he91" label="u:r:kernel:s0")
nov 29 17:50:11 Xperia10III mce[1996]: modules/audiorouting.c: audio_route_sink(): unknown audio sink device = 'bthfpforcall'
nov 29 17:50:11 Xperia10III mce[1996]: modules/audiorouting.c: audio_route_sink(): unknown audio sink device = 'bthfpforcall'
nov 29 17:50:11 Xperia10III kernel: sec_ts 1-0048: [sec_input] [P] tID:0 x:861 y:180 z:63 major:49 minor:41 tc:3 type:0 noise:0
nov 29 17:50:11 Xperia10III ofonod[3338]: Server: < AT+VGS=8\r
nov 29 17:50:11 Xperia10III ofonod[3338]: Server: > \r\nOK\r\n
One possibility could be that the headset is a bit flaky (which it certainly is) and triggers automatic re-routing as if it had just been connected.
A suggestion would be to disable all automatic switching, at least to Bluetooth, after the user has switched manually.