With a caveat: If you cross 32G heap size, the JVM will start using more and you effectively have less memory available until you get to 48G.

https://confluence.atlassian.com/jirakb/do-not-use-heap-size...

We had this problem for a while in one of our (non-Atlassian) systems and were running it with Xmx at 31.5G

So what your saying is your problem would have been solved if you just increased the number...