RMVJRNCHG

I have a performance problem with RMVJRNCHG command. What I'm trying to do,
is to run multiple parallel RMVJRNCHG commands for different subsets of
journaled objects in hope that parallel recovery is going to be faster than
serial. I never really expected parallel RMVJRNCHG to work, but, to my
surprise, it did. There was, however, a peculiarity. All parallel RMVJRNCHG
streams but one end in reasonable time and do not seem to interfere with
each other (no locks, etc.). The last job, however, takes hours and hours to
run; I even have an impression that it runs longer and accumulates more CPU
that a simple single-streamed RMVJRNCHG for the whole library. 

The collection of files that lags behind can differ from test to test. But
there is one thing they have in common:  the last RMVJRNCHG stream always
runs with NO FILES OPENED. 

In my view, it's a good candidate for an APAR. But did anybody see anything
like that? Does anybody have any suggestions apart from raising this with
IBM?

Lo