We believe it to be safe now.  In test run on multi-core machines it
is producing bit-for-bit identical result compared to the non-parallel
invocation.