'svnadmin load' and 'svnrdump load' both employ a hash which maps revisions
found in the dumpstream to actual committed revisions. It's been suggested that
this might not be the best approach in terms of memory consumption when loading
large (many-revisioned) repositories.
http://svn.haxx.se/dev/archive-2011-05/0943.shtml
A new approach *should* be able to be dropped in fairly easily. See the
set_revision_mapping() and get_revision_mapping() functions found in both
libsvn_repos/load-fs-vtable.c and svnrdump/load_editor.c.
Improvement
Major