Hi Benoît,
thanks for the truss output.
There are several known OS (Solaris) bugs regarding semtimedop with "Err#11 EAGAIN". Unfortunately we need the corresponding function calls (call stack) to locate the specific bug.
Please provide the function trace with DTrace or oradebug (i guess oradebug / short_stack should be enough in this case as well).
By the way .. are you running Oracle on a ZFS file system?
Regards
Stefan