I am looking for an open source workflow engine for my project. I have used Shark in several projects before. Shark is a heavyweight workflow engine and we have encountered some performance problems with it. So we want to switch to JBPM in my next project. But I found that JBPM does not use lazy fetching strategy because the lazy properties are set to false in most .hbm mapping files. I doubt this may also cause performance problems when there are lots of process instances in the database. It's possible that the whole database will be loaded into memory. Is there anyone has used it in a real project and can tell me is this the case when the process data grows.