What a ridiculous scenario. We had 2 news feeds pointing to the BBC website and they were causing the crash. I looked at the scheduled procs on the monitoring application and saw the exact point when it crashed was when the news aggregator updated.
I removed them both and the backup server worked fine as it passed through the proc. It's now been up for an hour after 10 minutes crashes.
Unbelievable really, there were 55,000 rows in the na_items table but only 2 on the na_sources table. Perhaps there was a dodgy entry on the site that was being downloaded?
Anyway, thanks for all of the help and we will work on the other recommendations for tweaking and fixing the duplicate email address error. That problem is caused by a single proc which attempts to grab a user id by email address (very wrong way of doing something!).
Cheers