We recently began a process to identify issues that may be slowing down our indexing process. One of the first things we were able to do was temporarily disable the MySQL query cache while the re-index was running.
The issue we now face is that Jira seems to chew through memory when running the indexing process. For what should be a relatively fixed throughput process, I believe Jira should not need a huge amount of heap.

We've found that when setting:
Xms4g
Xmx6g
, we start seeing constant Full GCs around the 50% indexed mark
When we set:
Xms6g
Xmx8g
, we see constant Full GCs around the 65% indexed mark
When we set:
Xms8g
Xmx10g
, we see constant Full GCs around the 78% indexed mark

We're using -XX:+UseParallelOldGC

Our instance size is as follows:
==========
Issues 446498
Projects 110
Custom Fields 573
Workflows 43
Attachments 263707
Comments 1733762
Users 8383
Groups 988
===========

We're seeing this behaviour on Jira 5.0.4