The issue has been fixed in the 11.2.0.2 patchset.
Possible workarounds for the 11.2.0.1 release are:
use the CLUSTER or SERVICE_NAME parameters to constrain the IMPDP job never to run on the instance not having access to the dump file
configure the database parameters in such a way that PQ/PX slaves are never started on the instance not having access to the dump file
use the PARALLEL=1 IMPDP command line parameter to disable parallel functionality during import
store all dump files on a local filesystem of one of the RAC nodes and use a parallel degree lower than the number of dump files and import with the CLUSTER=N command line option
加上cluster=N解决：
nohup impdp HQ_X1/HQ_X1 DUMPFILE=HQ_X1%u.DMP? transform=oid:n DIRECTORY=dir1 parallel=8 cluster=n&
相比于exp/imp来，expdp/impdp的BUG似乎有点多。