diff --git a/test_runner/performance/test_bulk_tenant_create.py b/test_runner/performance/test_bulk_tenant_create.py
index 9b05903cfa..e8daa96be6 100644
--- a/test_runner/performance/test_bulk_tenant_create.py
+++ b/test_runner/performance/test_bulk_tenant_create.py
@@ -1,7 +1,10 @@
+import threading
 import timeit
+from threading import BoundedSemaphore
 
 import pytest
 from fixtures.benchmark_fixture import MetricReport
+from fixtures.compare_fixtures import NeonCompare
 from fixtures.neon_fixtures import NeonEnvBuilder
 
 # Run bulk tenant creation test.
@@ -50,3 +53,57 @@ def test_bulk_tenant_create(
         "s",
         report=MetricReport.LOWER_IS_BETTER,
     )
+
+
+@pytest.mark.parametrize("tenants_count", [50])
+def test_parallel_tenant_create(
+    neon_compare: NeonCompare,
+    tenants_count: int,
+):
+    """Create lots of tenants in parallel
+
+    One important thing that this measures is the amount of prometheus
+    metrics per tenant. The pageserver exposes a lot of metrics for
+    each timeline, and this test gives some visibility to how much
+    exactly.  (We've had to raise the prometheus scraper's limit on
+    the max metrics size several times, because we expose so many.)
+    """
+    env = neon_compare.env
+    zenbenchmark = neon_compare.zenbenchmark
+
+    max_concurrent = 5
+    pool_sema = BoundedSemaphore(value=max_concurrent)
+
+    def worker(i: int):
+        with pool_sema:
+            tenant, timeline_id = env.neon_cli.create_tenant()
+
+            endpoint_tenant = env.endpoints.create_start("main", tenant_id=tenant)
+
+            with endpoint_tenant.cursor() as cur:
+                cur.execute("select count(*) from pg_class")
+                endpoint_tenant.stop()
+
+    threads = [threading.Thread(target=worker, args=(i,)) for i in range(tenants_count)]
+    start = timeit.default_timer()
+    for thread in threads:
+        thread.start()
+    for thread in threads:
+        thread.join()
+
+    end = timeit.default_timer()
+
+    zenbenchmark.record(
+        "tenant_creation_time",
+        end - start,
+        "s",
+        report=MetricReport.LOWER_IS_BETTER,
+    )
+
+    metrics = env.pageserver.http_client().get_metrics_str()
+    zenbenchmark.record(
+        "prometheus_metrics_size",
+        len(metrics) / tenants_count,
+        "bytes",
+        report=MetricReport.LOWER_IS_BETTER,
+    )
diff --git a/test_runner/regress/test_metric_collection.py b/test_runner/regress/test_metric_collection.py
index ecbce1f8f7..2f6f8ffe04 100644
--- a/test_runner/regress/test_metric_collection.py
+++ b/test_runner/regress/test_metric_collection.py
@@ -51,7 +51,7 @@ metric_kinds_checked = set([])
 
 
 #
-# verify that metrics look minilally sane
+# verify that metrics look minimally sane
 #
 def metrics_handler(request: Request) -> Response:
     if request.json is None: