diff --git a/core/libs/prx/libkernel/CMakeLists.txt b/core/libs/prx/libkernel/CMakeLists.txt
index 5e507ed890..e12d5a987f 100644
--- a/core/libs/prx/libkernel/CMakeLists.txt
+++ b/core/libs/prx/libkernel/CMakeLists.txt
@@ -2052,4 +2052,15 @@ if(BUILD_TESTING)
             endif()
         endforeach()
     endforeach()
+    foreach(sleepApi IN ITEMS sce_usleep sce_nanosleep sce_sleep)
+        set(sleepTest guest_sleep_cancel_${sleepApi}_deferred)
+        add_test(NAME ${sleepTest} COMMAND guest_sleep_cancel_tests deferred ${sleepApi})
+        add_test(NAME ${sleepTest}_coarse COMMAND guest_sleep_cancel_tests deferred ${sleepApi})
+        set_tests_properties(${sleepTest} ${sleepTest}_coarse PROPERTIES TIMEOUT 20)
+        set_tests_properties(${sleepTest}_coarse PROPERTIES ENVIRONMENT "APS5_COARSE_TIMED_WAITS=1")
+        if(WIN32)
+            set_tests_properties(${sleepTest} ${sleepTest}_coarse PROPERTIES
+                ENVIRONMENT_MODIFICATION "PATH=path_list_prepend:$<TARGET_FILE_DIR:libkernel>")
+        endif()
+    endforeach()
 endif()
diff --git a/core/libs/prx/libkernel/Time/TimedWait.cpp b/core/libs/prx/libkernel/Time/TimedWait.cpp
index bb5ba2cdca..b60cfb995d 100644
--- a/core/libs/prx/libkernel/Time/TimedWait.cpp
+++ b/core/libs/prx/libkernel/Time/TimedWait.cpp
@@ -619,6 +619,11 @@ void SleepNanos(std::uint64_t nanos) {
     }
     SleepUntil(NowNanos() + nanos);
 #else
+    struct SleepRegistration {
+        std::atomic<int>* state = WaitState().bound;
+        SleepRegistration() { state->fetch_add(1, std::memory_order_seq_cst); }
+        ~SleepRegistration() { state->fetch_sub(1, std::memory_order_seq_cst); }
+    } registration;
     struct timespec req{};
     req.tv_sec = static_cast<time_t>(nanos / 1000000000ULL);
     req.tv_nsec = static_cast<long>(nanos % 1000000000ULL);
diff --git a/core/libs/tests/GuestSleepCancel.cpp b/core/libs/tests/GuestSleepCancel.cpp
index 3b3ea39e9f..11daf21aac 100644
--- a/core/libs/tests/GuestSleepCancel.cpp
+++ b/core/libs/tests/GuestSleepCancel.cpp
@@ -39,9 +39,11 @@ static bool pending = false;
 static unsigned seconds = LongSleep;
 static std::string_view api = "sleep";
 static unsigned cleanupCalls = 0;
+static std::chrono::steady_clock::time_point sleepStart{};
 static std::chrono::steady_clock::duration elapsed{};
 
 static void APS5_VABI Cleanup(void*) {
+    elapsed = std::chrono::steady_clock::now() - sleepStart;
     ++cleanupCalls;
 }
 
@@ -64,9 +66,9 @@ static void* APS5_VABI Sleeper(void*) {
     ready.store(true);
     ready.notify_all();
     if (pending) release.wait(false);
-    const auto start = std::chrono::steady_clock::now();
+    sleepStart = std::chrono::steady_clock::now();
     Require(Sleep() == 0);
-    elapsed = std::chrono::steady_clock::now() - start;
+    elapsed = std::chrono::steady_clock::now() - sleepStart;
     returned.store(true);
     if (disabled) {
         Require(scePthreadSetcancelstate(0, nullptr) == 0);
@@ -86,13 +88,15 @@ static bool WaitForCompletion(Pthread thread) {
     return true;
 }
 
-static bool WaitForSleep(Pthread thread) {
+static bool WaitForSleep(Pthread thread, bool hostSleep) {
     const auto limit = std::chrono::steady_clock::now() + ObservationLimit;
     for (;;) {
         {
             std::lock_guard lock(thread->cancelLock);
             if (thread->cancelWait != nullptr) return true;
         }
+        if (hostSleep && thread->waitCount.load(std::memory_order_seq_cst) > 0) return true;
+        if (thread->_finished.load(std::memory_order_acquire)) return false;
         if (std::chrono::steady_clock::now() >= limit) return false;
         std::this_thread::yield();
     }
@@ -106,24 +110,33 @@ int main(int count, char** arguments) {
     const bool zero = mode == "pending";
     pending = zero || disabled || mode == "pending_long";
     const bool normal = mode == "normal";
-    Require(mode == "blocked" || disabled || pending || normal);
-    if (disabled || normal) seconds = 1;
+    const bool deferred = mode == "deferred";
+    Require(mode == "blocked" || disabled || pending || normal || deferred);
+    Require(!deferred || api == "sce_usleep" || api == "sce_nanosleep" || api == "sce_sleep");
+    if (disabled || normal || deferred) seconds = 1;
     if (zero) seconds = 0;
     Pthread thread = nullptr;
     Require(scePthreadCreate(&thread, nullptr, Sleeper, nullptr, nullptr) == 0);
-    ready.wait(false);
+    const auto readyLimit = std::chrono::steady_clock::now() + ObservationLimit;
+    while (!ready.load()) {
+        Require(std::chrono::steady_clock::now() < readyLimit);
+        std::this_thread::yield();
+    }
     bool wasSleeping = false;
     if (!normal) {
-        if (!pending) wasSleeping = WaitForSleep(thread);
+        if (!pending) {
+            wasSleeping = WaitForSleep(thread, deferred);
+            Require(wasSleeping);
+        }
         Require(scePthreadCancel(thread) == 0);
         release.store(true);
         release.notify_all();
     }
     const bool completed = WaitForCompletion(thread);
-    void* result = nullptr;
-    Require(scePthreadJoin(thread, &result) == 0);
     if (!completed) std::fputs("sleep stayed blocked after cancellation\n", stderr);
     Require(completed);
+    void* result = nullptr;
+    Require(scePthreadJoin(thread, &result) == 0);
     if (normal) {
         Require(result == nullptr && returned.load());
         Require(cleanupCalls == 0);
@@ -133,5 +146,5 @@ int main(int count, char** arguments) {
         Require(returned.load() == disabled);
         if (!pending) Require(wasSleeping);
     }
-    if (normal || disabled) Require(elapsed >= std::chrono::seconds(1));
+    if (normal || disabled || deferred) Require(elapsed >= std::chrono::seconds(1));
 }