diff --git a/tests/cloudpickle_test.py b/tests/cloudpickle_test.py
index 8bd210abe..2dca8abf0 100644
--- a/tests/cloudpickle_test.py
+++ b/tests/cloudpickle_test.py
@@ -1533,7 +1533,16 @@ def process_data():
             # grown by more than a few MB as closures are garbage collected at
             # the end of each remote function call.
             growth = w.memsize() - reference_size
-            assert growth < 1e7, growth
+
+            # For some reason, the memory growth after processing 100MB of
+            # data is ~10MB on MacOS, and ~1MB on Linux, so the upper bound on
+            # memory growth we use is only tight for MacOS. However,
+            # - 10MB is still 10x lower than the expected memory growth in case
+            # of a leak (which would be the total size of the processed data,
+            # 100MB)
+            # - the memory usage growth does not increase if using 10000
+            # iterations instead of 100 as used now (100x more data)
+            assert growth < 1.5e7, growth
 
         """.format(protocol=self.protocol)
         assert_run_python_script(code)