diff --git a/sparse_strips/vello_sparse_tests/tests/renderer.rs b/sparse_strips/vello_sparse_tests/tests/renderer.rs
index c610c1d..c3fb50d 100644
--- a/sparse_strips/vello_sparse_tests/tests/renderer.rs
+++ b/sparse_strips/vello_sparse_tests/tests/renderer.rs
@@ -216,7 +216,25 @@
         Self::set_transform(self, transform);
     }
 
+    // This method creates device resources every time it is called. This does not matter much for
+    // testing, but should not be used as a basis for implementing something real. This would be a
+    // very bad example for that.
     fn render_to_pixmap(&self, pixmap: &mut Pixmap, _: RenderMode) {
+        // On some platforms using `cargo test` triggers segmentation faults in wgpu when the GPU
+        // tests are run in parallel (likely related to the number of device resources being
+        // requested simultaneously). This is "fixed" by putting a mutex around this method,
+        // ensuring only one set of device resources is alive at the same time. This slows down
+        // testing when `cargo test` is used.
+        //
+        // Testing with `cargo nextest` (as on CI) is not meaningfully slowed down. `nextest` runs
+        // each test in its own process (<https://nexte.st/docs/design/why-process-per-test/>),
+        // meaning there is no contention on this mutex.
+        let _guard = {
+            use std::sync::Mutex;
+            static M: Mutex<()> = Mutex::new(());
+            M.lock().unwrap()
+        };
+
         let width = self.width();
         let height = self.height();