diff --git a/CMakeLists.txt b/CMakeLists.txt
index 3d9e9f1..783d14a 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -2329,9 +2329,9 @@
     )
 
     if(NOT WIN32 AND ${REACTOR_BACKEND} STREQUAL "Subzero")
-        target_link_libraries(ReactorUnitTests ${Reactor} marl pthread dl)
+        target_link_libraries(ReactorUnitTests ${Reactor} pthread dl)
     else()
-        target_link_libraries(ReactorUnitTests ${Reactor} marl)
+        target_link_libraries(ReactorUnitTests ${Reactor})
     endif()
 
     set(GLES_UNITTESTS_LIST
diff --git a/src/Reactor/ReactorUnitTests.cpp b/src/Reactor/ReactorUnitTests.cpp
index 7e38d93..d5aafa9 100644
--- a/src/Reactor/ReactorUnitTests.cpp
+++ b/src/Reactor/ReactorUnitTests.cpp
@@ -18,9 +18,6 @@
 
 #include "gtest/gtest.h"
 
-#include "marl/defer.h"
-#include "marl/scheduler.h"
-
 #include <array>
 #include <cmath>
 #include <thread>
@@ -2137,11 +2134,6 @@
 		return;
 	}
 
-	marl::Scheduler scheduler;
-	scheduler.setWorkerThreadCount(8);
-	scheduler.bind();
-	defer(scheduler.unbind());
-
 	Coroutine<int()> function;
 	{
 		Yield(Int(0));
@@ -2175,11 +2167,6 @@
 		return;
 	}
 
-	marl::Scheduler scheduler;
-	scheduler.setWorkerThreadCount(8);
-	scheduler.bind();
-	defer(scheduler.unbind());
-
 	Coroutine<uint8_t(uint8_t * data, int count)> function;
 	{
 		Pointer<Byte> data = function.Arg<0>();
@@ -2221,11 +2208,6 @@
 		return;
 	}
 
-	marl::Scheduler scheduler;
-	scheduler.setWorkerThreadCount(8);
-	scheduler.bind();
-	defer(scheduler.unbind());
-
 	Coroutine<int()> function;
 	{
 		Int4 a{ 1, 2, 3, 4 };
@@ -2260,11 +2242,6 @@
 		return;
 	}
 
-	marl::Scheduler scheduler;
-	scheduler.setWorkerThreadCount(8);
-	scheduler.bind();
-	defer(scheduler.unbind());
-
 	for(int i = 0; i < 2; ++i)
 	{
 		Coroutine<int()> function;
@@ -2289,11 +2266,6 @@
 		return;
 	}
 
-	marl::Scheduler scheduler;
-	scheduler.setWorkerThreadCount(8);
-	//scheduler.bind();
-	//defer(scheduler.unbind());
-
 	Coroutine<int()> function;
 	{
 		Yield(Int(0));
@@ -2318,9 +2290,6 @@
 	for(size_t t = 0; t < numThreads; ++t)
 	{
 		threads.emplace_back([&] {
-			scheduler.bind();
-			defer(scheduler.unbind());
-
 			auto coroutine = function();
 
 			for(size_t i = 0; i < fibonacci.size(); i++)
diff --git a/src/Reactor/SubzeroReactor.cpp b/src/Reactor/SubzeroReactor.cpp
index df12cb1..9a83795 100644
--- a/src/Reactor/SubzeroReactor.cpp
+++ b/src/Reactor/SubzeroReactor.cpp
@@ -268,7 +268,16 @@
 // Coroutine globals
 rr::Type *coroYieldType = nullptr;
 std::shared_ptr<rr::CoroutineGenerator> coroGen;
+marl::Scheduler &getOrCreateScheduler()
+{
+	static auto scheduler = [] {
+		auto s = std::make_unique<marl::Scheduler>();
+		s->setWorkerThreadCount(8);
+		return s;
+	}();
 
+	return *scheduler;
+}
 }  // Anonymous namespace
 
 namespace {
@@ -4467,6 +4476,7 @@
 // Lifetime: from yield to when CoroutineEntryDestroy generated function is called.
 struct CoroutineData
 {
+	bool useInternalScheduler = false;
 	marl::Event suspended;                                // the coroutine is suspended on a yield()
 	marl::Event resumed;                                  // the caller is suspended on an await()
 	marl::Event done{ marl::Event::Mode::Manual };        // the coroutine should stop at the next yield()
@@ -4510,9 +4520,13 @@
 void stop(Nucleus::CoroutineHandle handle)
 {
 	auto *coroData = reinterpret_cast<CoroutineData *>(handle);
-	coroData->done.signal();               // signal that the coroutine should stop at next (or current) yield.
-	coroData->resumed.signal();            // wake the coroutine if blocked on a yield.
-	coroData->terminated.wait();           // wait for the coroutine to return.
+	coroData->done.signal();      // signal that the coroutine should stop at next (or current) yield.
+	coroData->resumed.signal();   // wake the coroutine if blocked on a yield.
+	coroData->terminated.wait();  // wait for the coroutine to return.
+	if(coroData->useInternalScheduler)
+	{
+		::getOrCreateScheduler().unbind();
+	}
 	coro::destroyCoroutineData(coroData);  // free the coroutine data.
 }
 
@@ -4752,6 +4766,12 @@
 	// This doubles up as our coroutine handle
 	auto coroData = coro::createCoroutineData();
 
+	coroData->useInternalScheduler = (marl::Scheduler::get() == nullptr);
+	if(coroData->useInternalScheduler)
+	{
+		::getOrCreateScheduler().bind();
+	}
+
 	marl::schedule([=] {
 		// Store handle in TLS so that the coroutine can grab it right away, before
 		// any fiber switch occurs.