I think this is worth investigating indeed. Have we got a redmine issue on it yet? I guess those would be two: a) find out the source of SGSN errors and b) be more tolerant of errors in the GGSN.
Possibly b) becomes obsolete when a) is resolved, or maybe b) is even harmful when errors should have implied the end of a PDP context.
For a), it would be super interesting to get a pcap / log snippets of the actual error indications attached to the redmine issue.
I've updated the redmine issue with the relevant packet capture.
https://osmocom.org/issues/3696
Regards,
Pierre