Per default, the scheduled procedure
runs every 10 minutes. That explains the repeating behavior.
concerning the crash: it looks like your server is running out of addressable memory per process (typically 2GB for 32bit processes). All threads share the same address space of the process. If the process runs close to 2GB, and a thread needs more memory, the OS says "no" and kills it. Running out of this limit can have multiple origins; often, calls to external programs can lead to this problem. Adding more physical memory only helps against running out of memory on the OS-level, including swap (rather uncommon, depends what is running aside openacs on the system), but it does not change the 2GB limit. To overcome it, one could compile the server + tcl + modules with 64bit, but in your case, i would not recommend it, since it won't address the cause of the problem.
In your case, i have the following suspicion: Since you wrote that your system has been modified to accept duplicate email addresses a while ago, i would expect that the delivery of notifications is broken since this time. This means that, since this time, notifications are piling up. If there is a huge number of notifications waiting for delivery, and these are loaded into memory, you might run out of resources.
Check in psql how many requests are waiting:
select count(*) from notifications
The number should be low, typically 0.
To understand the notification sweeper, check:
Hope, this helps a little.