I've spoken with Juergen Hoeller about the Spring aspect of this and his feedback was:
Anyway, the issue itself looks rather odd. Spring's JtaTransactionManager just delegates to UserTransaction.begin/commit/rollback, and checks that UserTransaction's status first, so certainly wouldn't ignore any existing transactions there. I rather suspect that the reporter had an additional transaction management strategy involved next to JtaTransactionManager, e.g. a Spring DataSourceTransactionManager which does obtain and commit JDBC Connections independently.
I suggest that if this is still an issue that you follow up with Spring via the relevant forum.
I'm not against adding the requested feature but it isn't something I am likely to work on myself. As always, patches welcome. Without a patch this issue will eventually get resolved as WONTFIX.