The issue of the jvm crashing has been an elusive one so I am going to try to reproduce it using the 1.2 code base. Starting with 4 x 1.5 hour captures with 1 minute in between. If 1.2 doesn't have the issue then something we have changed between 1.2 and 1.3 has broken the codebase. There is a torrent of commits that effected the capture agent in 1.3 including a lot of merges from other branches and trunk to 1.3 meaning these will need to be tracked down one by one individually (can't be done with a collective diff).
6 x 1.5 hours with 1 minute gaps so far successful.
All 12 captures @ 1.5 hours with 1 minute gaps finished successfully. This is a good indication that 1.2 doesn't suffer from the issue.