Sergio beat me on time and announced Scarlet beta 2
Well, at least this delay gives me more time to comment on some blog reactions from Terracotta friends Orion and Ari :)
As Ari said, clustering with an invasive situation is really a different beast. I'd add that it could be literally devastating for your legacy project. Atlassian Confluence is a real world example, it has been a 54 man/month effort according to an atlassian presentation given some months ago. Clustering Jira with Terracotta has been much easier than that, and assuming that confluence and jira have the same complexity (which is a reasonable assumption, coming from the same company and sharing a similar architecture), we can safely say that the effort to cluster with TC has been a tiny fraction than 54 man/month, so much much better than that.
Regarding the transparent clustering claim, well... for us it's been more a translucid one ;) Dealing with legacy J2ee code can be very tricky, and using a straightforward terracotta solution (ie simply putting the objects in your shared roots and writing some XML) is not always feasible/performant/possible. We described some of the issues we found in our talk at javaday (which should be replicated at next javapolis, if we can find the air tickets :) ).
For example, dealing with a lot of uncorrectly synchronized home-grown caches forced us to write a common cache adapter on top of them, using a lock-striping solution and than TC-ize that.
Not exactly the kind of stuff you code in the weekend!