diff --git a/benchmarks/core/object_db.gd b/benchmarks/core/object_db.gd
new file mode 100644
index 0000000..fc129f7
--- /dev/null
+++ b/benchmarks/core/object_db.gd
@@ -0,0 +1,113 @@
+extends Benchmark
+
+const NUM_OBJECTS := 1000
+const ITERATIONS := 100
+
+var threads: Array[Thread]
+var sem := Semaphore.new()
+
+func _init_threads(p_count: int, p_contention_min: float = 1.0, p_contention_max: float = 1.0):
+    var contention := p_contention_max
+    for i in p_count:
+        var th := Thread.new()
+        var duty: float
+        if i == 0:
+            duty = 1.0
+        else:
+            # Min is never reached, so 0 can be used without the last thread having 0 duties.
+            duty = remap(i + 1, 0, p_count, p_contention_min, p_contention_max)
+        th.start(_thread_func.bind(duty))
+        threads.push_back(th)
+    sem.post(threads.size())
+
+func _term_threads():
+    for th in threads:
+        th.wait_to_finish()
+    threads = []
+
+func _thread_func(p_duty: float) -> void:
+    var objs: Array[Object]
+    objs.resize(NUM_OBJECTS)
+    sem.wait()
+    for i in int(ITERATIONS * p_duty):
+        for j in NUM_OBJECTS:
+            objs[j] = Object.new()
+        for j in NUM_OBJECTS:
+            objs[j].free()
+
+
+func benchmark_single() -> void:
+    _init_threads(1)
+    _term_threads()
+
+
+# In the benchmarks below, different contention terms are used:
+# full - All threads run all the iterations.
+# half - Threads other than the first one run half of the iterations.
+# slope - Threads other than the first one run successively more work,
+#         with the second thread running few iterations and
+#         the last thread running almost as many as the first one.
+# little - Threads other than the first one run few iterations.
+
+func benchmark_2_threads_full_contention() -> void:
+    _init_threads(2)
+    _term_threads()
+
+func benchmark_2_threads_half_contention() -> void:
+    _init_threads(2, 0.5, 0.5)
+    _term_threads()
+
+func benchmark_2_threads_little_contention() -> void:
+    _init_threads(2, 0.1, 0.1)
+    _term_threads()
+
+
+func benchmark_4_threads_full_contention() -> void:
+    _init_threads(4)
+    _term_threads()
+
+func benchmark_4_threads_half_contention() -> void:
+    _init_threads(4, 0.5, 0.5)
+    _term_threads()
+
+func benchmark_4_threads_slope_contention() -> void:
+    _init_threads(4, 0.0, 1.0)
+    _term_threads()
+
+func benchmark_4_threads_little_contention() -> void:
+    _init_threads(4, 0.1, 0.1)
+    _term_threads()
+
+
+func benchmark_8_threads_full_contention() -> void:
+    _init_threads(8)
+    _term_threads()
+
+func benchmark_8_threads_half_contention() -> void:
+    _init_threads(8, 0.5, 0.5)
+    _term_threads()
+
+func benchmark_8_threads_slope_contention() -> void:
+    _init_threads(8, 0.0, 1.0)
+    _term_threads()
+
+func benchmark_8_threads_little_contention() -> void:
+    _init_threads(8, 0.1, 0.1)
+    _term_threads()
+
+
+func benchmark_12_threads_full_contention() -> void:
+    _init_threads(12)
+    _term_threads()
+
+func benchmark_12_threads_half_contention() -> void:
+    _init_threads(12, 0.5, 0.5)
+    _term_threads()
+
+func benchmark_12_threads_slope_contention() -> void:
+    _init_threads(12, 0.0, 1.0)
+    _term_threads()
+
+func benchmark_12_threads_little_contention() -> void:
+    _init_threads(12, 0.1, 0.1)
+    _term_threads()