diff --git a/libraries/AP_HAL_SITL/SITL_State.cpp b/libraries/AP_HAL_SITL/SITL_State.cpp
index 534df20252..72b73d2b5a 100644
--- a/libraries/AP_HAL_SITL/SITL_State.cpp
+++ b/libraries/AP_HAL_SITL/SITL_State.cpp
@@ -186,14 +186,19 @@ void SITL_State::wait_clock(uint64_t wait_time_usec)
             _fdm_input_step();
         } else {
 #ifdef CYGWIN_BUILD
-            if (speedup > 2) {
-                // this effectively does a yield of the CPU. The
-                // granularity of sleeps on cygwin is very high, so
-                // this is needed for good thread performance. We
-                // don't do this at low speedups as it causes the cpu
-                // to run hot
-                usleep(0);
-                continue;
+            if (speedup > 2 && hal.util->get_soft_armed()) {
+                const char *current_thread = Scheduler::from(hal.scheduler)->get_current_thread_name();
+                if (current_thread && strcmp(current_thread, "Scripting") == 0) {
+                    // this effectively does a yield of the CPU. The
+                    // granularity of sleeps on cygwin is very high,
+                    // so this is needed for good thread performance
+                    // in scripting. We don't do this at low speedups
+                    // as it causes the cpu to run hot
+                    // We also don't do it while disarmed, as lua performance is less
+                    // critical while disarmed
+                    usleep(0);
+                    continue;
+                }
             }
 #endif
             usleep(1000);
diff --git a/libraries/AP_HAL_SITL/Scheduler.cpp b/libraries/AP_HAL_SITL/Scheduler.cpp
index a1cbd8b8a9..ba04d6af2d 100644
--- a/libraries/AP_HAL_SITL/Scheduler.cpp
+++ b/libraries/AP_HAL_SITL/Scheduler.cpp
@@ -308,6 +308,7 @@ void Scheduler::stop_clock(uint64_t time_usec)
 void *Scheduler::thread_create_trampoline(void *ctx)
 {
     struct thread_attr *a = (struct thread_attr *)ctx;
+    a->thread = pthread_self();
     a->f[0]();
     
     WITH_SEMAPHORE(_thread_sem);
@@ -408,3 +409,15 @@ void Scheduler::check_thread_stacks(void)
         }
     }
 }
+
+// get the name of the current thread, or nullptr if not known
+const char *Scheduler::get_current_thread_name(void) const
+{
+    const pthread_t self = pthread_self();
+    for (struct thread_attr *a=threads; a; a=a->next) {
+        if (a->thread == self) {
+            return a->name;
+        }
+    }
+    return nullptr;
+}
diff --git a/libraries/AP_HAL_SITL/Scheduler.h b/libraries/AP_HAL_SITL/Scheduler.h
index e17c138884..518cad42a3 100644
--- a/libraries/AP_HAL_SITL/Scheduler.h
+++ b/libraries/AP_HAL_SITL/Scheduler.h
@@ -67,6 +67,9 @@ public:
     // a couple of helper functions to cope with SITL's time stepping
     bool semaphore_wait_hack_required() const;
 
+    // get the name of the current thread, or nullptr if not known
+    const char *get_current_thread_name(void) const;
+
 private:
     SITL_State *_sitlState;
     uint8_t _nested_atomic_ctr;
@@ -105,6 +108,7 @@ private:
         void *stack;
         const uint8_t *stack_min;
         const char *name;
+        pthread_t thread;
     };
     static struct thread_attr *threads;
     static const uint8_t stackfill = 0xEB;