diff --git a/tests/v1/kv_offload/tiering/test_obj_tier.py b/tests/v1/kv_offload/tiering/test_obj_tier.py
index 2dba5baf3d6f..640848bda6f9 100644
--- a/tests/v1/kv_offload/tiering/test_obj_tier.py
+++ b/tests/v1/kv_offload/tiering/test_obj_tier.py
@@ -54,6 +54,7 @@ def _make_offloading_config(
     world_size: int | None = None,
     replicated_layout: bool = False,
     is_parallelism_agnostic: bool = False,
+    blocks_per_chunk: int = 1,
 ) -> OffloadingConfig:
     if world_size is None:
         world_size = tp_size
@@ -64,7 +65,9 @@ def _make_offloading_config(
         extra_config={},
         engine_id="test-engine",
         model=OffloadingModelConfig(name="test/model", dtype="float16"),
-        cache=OffloadingCacheConfig(tokens_per_hash=16, blocks_per_chunk=1),
+        cache=OffloadingCacheConfig(
+            tokens_per_hash=16, blocks_per_chunk=blocks_per_chunk
+        ),
         parallel=OffloadingParallelConfig(
             rank=rank,
             world_size=world_size,
@@ -83,6 +86,7 @@ def _make_offloading_config(
 
 _OFFLOADING_SPEC = SimpleNamespace(
     config=_make_offloading_config(enable_kv_cache_events=False),
+    blocks_per_chunk=1,
 )
 
 _STORE_CONFIG = {
@@ -221,6 +225,7 @@ def _make_events_spec(enable_kv_cache_events: bool) -> SimpleNamespace:
     """Offloading spec stub with an explicit global KV events flag."""
     return SimpleNamespace(
         config=_make_offloading_config(enable_kv_cache_events),
+        blocks_per_chunk=1,
         kv_events_config=OffloadingKVEventsConfig(
             enable_kv_cache_events=enable_kv_cache_events,
             self_describing_kv_events=False,
@@ -780,6 +785,7 @@ def test_obj_tier_replicated_layout_collapses_mapper_identity():
         config=_make_offloading_config(
             False, tp_size=2, world_size=2, rank=1, replicated_layout=True
         ),
+        blocks_per_chunk=1,
         kv_events_config=OffloadingKVEventsConfig(
             enable_kv_cache_events=False,
             self_describing_kv_events=False,
@@ -789,6 +795,7 @@ def test_obj_tier_replicated_layout_collapses_mapper_identity():
         config=_make_offloading_config(
             False, tp_size=4, world_size=4, rank=3, replicated_layout=True
         ),
+        blocks_per_chunk=1,
         kv_events_config=OffloadingKVEventsConfig(
             enable_kv_cache_events=False,
             self_describing_kv_events=False,
@@ -806,3 +813,22 @@ def test_obj_tier_replicated_layout_collapses_mapper_identity():
     finally:
         tp2_tier.shutdown()
         tp4_tier.shutdown()
+
+
+def test_obj_tier_namespace_includes_blocks_per_chunk():
+    """Chunks of different sizes have different layouts, so they must not share
+    object keys (the fs tier namespaces by chunk size the same way)."""
+    tiers = []
+    for blocks_per_chunk in (1, 4):
+        spec = SimpleNamespace(
+            config=_make_offloading_config(False, blocks_per_chunk=blocks_per_chunk),
+            blocks_per_chunk=blocks_per_chunk,
+        )
+        tiers.append(_make_tier(offloading_spec=spec)[0])
+    try:
+        assert tiers[0]._file_mapper.get_file_name(key(1)) != (
+            tiers[1]._file_mapper.get_file_name(key(1))
+        )
+    finally:
+        for tier in tiers:
+            tier.shutdown()
diff --git a/vllm/v1/kv_offload/tiering/obj/manager.py b/vllm/v1/kv_offload/tiering/obj/manager.py
index 26e78894742a..37ce4ca3e244 100644
--- a/vllm/v1/kv_offload/tiering/obj/manager.py
+++ b/vllm/v1/kv_offload/tiering/obj/manager.py
@@ -170,7 +170,10 @@ def __init__(
         root_dir = f"{prefix}/" if prefix else ""
         # Opt in; FileMapper enables it only for a parallelism-invariant block.
         self._file_mapper = FileMapper.from_offloading_spec(
-            root_dir, offloading_spec, parallel_agnostic=True
+            root_dir,
+            offloading_spec,
+            blocks_per_file=offloading_spec.blocks_per_chunk,
+            parallel_agnostic=True,
         )
         self._next_obj_dev_id: int = 1  # dev_id=0 is reserved for _exists() probes