diff --git a/swh/loader/tar/build.py b/swh/loader/tar/build.py
--- a/swh/loader/tar/build.py
+++ b/swh/loader/tar/build.py
@@ -3,9 +3,16 @@
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
 
+import copy
+import logging
+import os
+
 import arrow
 
 
+logger = logging.getLogger(__name__)
+
+
 # Static setup
 EPOCH = 0
 UTC_OFFSET = 0
@@ -74,3 +81,28 @@
         'message': REVISION_MESSAGE,
         'synthetic': True,
     }
+
+
+def set_original_artifact(*, revision, filepath, nature, hashes):
+    """Set the original artifact data on the given revision for
+        the tarball currently being loaded."""
+
+    revision = copy.deepcopy(revision)
+    if 'metadata' not in revision or not revision['metadata']:
+        revision['metadata'] = {}
+    if 'original_artifact' in revision['metadata']:
+        oa = revision['metadata']['original_artifact']
+        if oa:
+            logger.warning(
+                'Revision already contains original_artifact metadata, '
+                'replacing: %r',
+                oa,
+            )
+
+    revision['metadata']['original_artifact'] = [{
+        'name': os.path.basename(filepath),
+        'archive_type': nature,
+        **hashes,
+    }]
+
+    return revision
diff --git a/swh/loader/tar/loader.py b/swh/loader/tar/loader.py
--- a/swh/loader/tar/loader.py
+++ b/swh/loader/tar/loader.py
@@ -18,7 +18,7 @@
 from swh.model.hashutil import MultiHash, HASH_BLOCK_SIZE
 from swh.model.from_disk import Directory
 
-from .build import compute_revision
+from .build import compute_revision, set_original_artifact
 
 try:
     from _version import __version__
@@ -263,16 +263,12 @@
         build the revision.
 
         """
-        return {
-            **compute_revision(filepath, self.last_modified),
-            'metadata': {
-                'original_artifact': [{
-                    'name': os.path.basename(filepath),
-                    'archive_type': nature,
-                    **hashes,
-                }],
-            }
-        }
+        return set_original_artifact(
+            revision=compute_revision(filepath, self.last_modified),
+            filepath=filepath,
+            nature=nature,
+            hashes=hashes,
+        )
 
     def build_snapshot(self, revision):
         """Build the snapshot targeting the revision.
@@ -325,16 +321,12 @@
         revision.
 
         """
-        return {
-            **self.revision,
-            'metadata': {
-                'original_artifact': [{
-                    'name': os.path.basename(filepath),
-                    'archive_type': nature,
-                    **hashes,
-                }],
-            }
-        }
+        return set_original_artifact(
+            revision=self.revision,
+            filepath=filepath,
+            nature=nature,
+            hashes=hashes,
+        )
 
     def build_snapshot(self, revision):
         """Build the snapshot targeting the revision.
diff --git a/swh/loader/tar/tests/test_loader.py b/swh/loader/tar/tests/test_loader.py
--- a/swh/loader/tar/tests/test_loader.py
+++ b/swh/loader/tar/tests/test_loader.py
@@ -85,6 +85,8 @@
         self.assertCountReleases(0)
         self.assertCountSnapshots(1)
 
+        return actual_revision
+
 
 class TestRemoteTarLoader(PrepareDataForTestLoader):
     """Test the remote loader scenario (local/remote)
@@ -194,7 +196,6 @@
 
     def test_load(self):
         """Load a local tarball should result in persisted swh data
-
         """
         # given
         origin = {
@@ -226,6 +227,10 @@
             'type': revision_type,
             'message': revision_message,
             'synthetic': True,
+            'metadata': {
+                'foo': 'bar',
+                'original_artifact': ['bogus_original_artifact'],
+            }
         }
 
         branch_name = os.path.basename(self.tarpath)
@@ -236,4 +241,10 @@
                          branch_name=branch_name)
 
         # then
-        self.assert_data_ok()
+        actual_revision = self.assert_data_ok()
+
+        # Check metadata passthrough
+        assert actual_revision['metadata']['foo'] == 'bar'
+
+        # FIXME: use the caplog pytest fixture to check that the clobbering of
+        # original artifact sent a warning