UPDATE
Its been about 24 hours since my last major change. I had set my /etc/default/graylog-server at half my available memory per the site instructions. Having avoided the issues I had been experiencing recently on an ElasticSearch-backed Graylog instance, before, with much less memory defined in the /etc/default/graylog-server, I backed it down to -Xms16g -Xmx16g on my current build. So far so good.