diff --git a/.github/workflows/tests.yaml b/.github/workflows/tests.yaml
index aebc97c0..03c0a02c 100644
--- a/.github/workflows/tests.yaml
+++ b/.github/workflows/tests.yaml
@@ -9,12 +9,17 @@ on:
 jobs:
   hal-cgp-testsuite:
     name: hal-cgp testsuite
-    runs-on: ubuntu-latest
+    runs-on: ${{ matrix.os }}
     strategy:
       matrix:
+        os: [ubuntu-latest, macos-11]
         python-version: [3.7, 3.8, 3.9]
         dep: ['[all]', '[dev]']
         exclude:
+          - os: macos-11
+            python-version: 3.7
+          - os: macos-11
+            python-version: 3.8
           - python-version: 3.7
             dep: "[dev]"
           - python-version: 3.8
diff --git a/test/test_utils.py b/test/test_utils.py
index 2b7ae5c8..5f4bfb2e 100644
--- a/test/test_utils.py
+++ b/test/test_utils.py
@@ -264,12 +264,12 @@ def evaluate_objective_on_list(x):
     assert (time.time() - t0) > (0.9 * sleep_time)
 
     # second call should be faster as result is retrieved from cache;
-    # at most 40% of the sleep time; to account for possible timing
+    # at most 50% of the sleep time; to account for possible timing
     # measurement inaccuracies and process spin up/down time in
     # TravisCI we should not choose less
     t0 = time.time()
     evaluate_objective_on_list(x)
-    assert (time.time() - t0) < (0.4 * sleep_time)
+    assert (time.time() - t0) < (0.5 * sleep_time)
 
 
 def test_cache_decorator_consistency(individual):