This is not easy to reproduce with a fresh XE. It probably require lot's of data to be noticeable.
I my case I reproduce it using the xwiki.org farm database the following way:
- delete the lucene index on fs
- start xwiki
- open an page so that lucene plugin start indexing
The memory consumed by the JVM quickly grows until it explode the limit allocated to it (about 750m accorindg to xinit.cfg) and produce a OOM.
Compared to 3.1.1 used with the exact same database on the same server:
- 3.1.1 stay stable around 350M memory according to top during the whole indexing process
- 3.2/3.3 quickly go up to 800M when the indexing is started
Of course they both have the same configuration in xwiki.cfg and xwiki.properties (except for new things).
The total size of the lucene index produced by 3.1.1 on fs is about 65M so I don't think it's an issue like the full index of Lucene kept in memory but probably more related to the document coming from the database. The configured document cache size is 2000 which does not sounds that much to me.