As far as I know, I don't think we have the mechanism of disabling persistence of large variables. You can file a request task in our jira system if you want.
One option may be to use an 'in-memory' process - see http://ode.apache.org/user-guide.html#UserGuide-Inmemoryexecution
However note the restriction - it only works if the process has a single receive, so effectively the process receives the message, does some work, and then completes. In this situation they don't use persistence.
From the description of your task, it sounds like this is a possibility, if the process receives a message, asks web service 1 for the large file, then upon receipt immediately sends it on to the next web service 2.
Might be worth a try.
Thanks Gary, that works for me - I can live with those constraints. The database did not increase appreciably even after sending through a 20MB file.