diff --git a/marimo/_plugins/ui/_impl/tables/narwhals_table.py b/marimo/_plugins/ui/_impl/tables/narwhals_table.py
index a2a82dc5ed8..e60c1f6132b 100644
--- a/marimo/_plugins/ui/_impl/tables/narwhals_table.py
+++ b/marimo/_plugins/ui/_impl/tables/narwhals_table.py
@@ -699,6 +699,10 @@ def get_sample_values(self, column: str) -> list[str | int | float]:
 
         # Sample 3 values from the column
         SAMPLE_SIZE = 3
+        # Cap nested list/dict serialization depth. Unbounded recursion (or
+        # even json.dumps on pathological nested structs) can stall dataset
+        # registration for deeply nested Polars Struct columns (#9378).
+        MAX_NESTING_DEPTH = 5
         try:
 
             def _json_default(o: Any) -> str:
@@ -706,6 +710,26 @@ def _json_default(o: Any) -> str:
                     return o.name
                 return str(o)
 
+            def _with_depth_limit(value: Any, depth: int = 0) -> Any:
+                """Return a structure safe for json.dumps / display.
+
+                At max depth, emit a fixed sentinel rather than str(value):
+                str() on large list/dict trees still walks the full remaining
+                container (issue #9378 branching payloads).
+                """
+                if depth >= MAX_NESTING_DEPTH:
+                    if isinstance(value, (list, dict)):
+                        return "..."
+                    return value
+                if isinstance(value, list):
+                    return [_with_depth_limit(v, depth + 1) for v in value]
+                if isinstance(value, dict):
+                    return {
+                        k: _with_depth_limit(v, depth + 1)
+                        for k, v in value.items()
+                    }
+                return value
+
             def to_primitive(value: Any) -> str | int | float:
                 if isinstance(value, Enum):
                     return value.name
@@ -713,7 +737,8 @@ def to_primitive(value: Any) -> str | int | float:
                     return value
                 if isinstance(value, (list, dict)):
                     try:
-                        return json.dumps(value, default=_json_default)
+                        limited = _with_depth_limit(value)
+                        return json.dumps(limited, default=_json_default)
                     except (TypeError, ValueError):
                         return str(value)
                 return str(value)
diff --git a/tests/_plugins/ui/_impl/tables/test_narwhals.py b/tests/_plugins/ui/_impl/tables/test_narwhals.py
index 53a3c222a4e..2611330f55f 100644
--- a/tests/_plugins/ui/_impl/tables/test_narwhals.py
+++ b/tests/_plugins/ui/_impl/tables/test_narwhals.py
@@ -1578,6 +1578,37 @@ def build_payload(row_idx: int, depth: int):
 
 
 @pytest.mark.skipif(not HAS_DEPS, reason="polars not installed")
+@pytest.mark.skipif(not HAS_DEPS, reason="polars not installed")
+def test_get_sample_values_branching_payload_is_bounded() -> None:
+    """Wide branching nests must not explode work via str() at the depth cap."""
+    import polars as pl
+
+    def branch(depth: int, width: int = 8) -> Any:
+        if depth == 0:
+            return {"leaf": "x" * 20, "n": 1}
+        return {f"b{i}": branch(depth - 1, width) for i in range(width)}
+
+    # width^depth without a real cap is enormous; with sentinel cap work is O(width^MAX).
+    payload = branch(6, width=6)
+    df = pl.DataFrame({"payload": [payload, payload, payload]})
+    manager = NarwhalsTableManager.from_dataframe(
+        nw.from_native(df, eager_only=True)
+    )
+
+    t0 = time.perf_counter()
+    samples = manager.get_sample_values("payload")
+    elapsed = time.perf_counter() - t0
+
+    assert len(samples) == 3
+    assert elapsed < 1.0, (
+        f"get_sample_values took {elapsed:.2f}s (expected < 1s)"
+    )
+    for s in samples:
+        assert isinstance(s, str)
+        assert "..." in s
+        assert len(s) < 250_000
+
+
 def test_get_sample_values_nested_struct_is_json() -> None:
     import polars as pl
 
@@ -1730,6 +1761,32 @@ def test_calculate_top_k_rows_with_all_special_floats(df: Any) -> None:
         ]
 
 
+@pytest.mark.skipif(
+    not DependencyManager.pandas.has(),
+    reason="pandas required for nested sample depth test",
+)
+def test_get_sample_values_nested_depth_limit() -> None:
+    """Deeply nested list/dict samples must finish quickly (#9378 / #9383)."""
+    import pandas as pd
+
+    def nest(depth: int) -> object:
+        v: object = "leaf"
+        for _ in range(depth):
+            v = {"k": v}
+        return v
+
+    df = pd.DataFrame({"deep": [nest(20)]})
+    manager = NarwhalsTableManager.from_dataframe(df)
+    t0 = time.perf_counter()
+    sample = manager.get_sample_values("deep")
+    elapsed = time.perf_counter() - t0
+    assert elapsed < 1.0, f"nested sample too slow: {elapsed:.3f}s"
+    assert len(sample) == 1
+    assert isinstance(sample[0], str)
+    # depth limit should stringify remaining nesting rather than exploding
+    assert "leaf" in sample[0] or sample[0].startswith("{")
+
+
 def _normalize_result(result: list[tuple[Any, int]]) -> list[tuple[Any, int]]:
     """Normalize None and NaN values for comparison."""
     out: list[tuple[Any, int]] = []