Currently kpropd saves the dump file being transmitted, then it loads
it.  It seems very likely that piping the dump straight to kdb5_util
load should improve full resync performance, though on 1 CPU systems
it could also backfire (by requiring many more context switches).

An option to pipe the dump straight to kdb5_util load seems
appropriate and desirable.  I propose using either an empty dump file
name or '-' to select this feature.