I see you have long-query logging enabled, what was this query doing? It seems like the oddball from your ps output, it was taking half your CPU. Or did you have to kill the db before it logged anything out. If so, while debugging something with a memory problem here we set up a cronjob to log out all running queries every minute, before the oom-killer would start killing stuff, maybe you could catch the culprit next time.
For the pids that you included strace outputs for, curious, why did you pick them? Were they using a lot of CPU? Except for the postmaster one (81372), I didn't see them in the ps output.
