diff --git a/fidget/src/mesh/mt/pool.rs b/fidget/src/mesh/mt/pool.rs
index e4f5a8c9..767bf313 100644
--- a/fidget/src/mesh/mt/pool.rs
+++ b/fidget/src/mesh/mt/pool.rs
@@ -56,7 +56,6 @@ impl ThreadPool {
             threads,
             counter: &self.counter,
             index,
-            phase: 1,
         }
     }
 }
@@ -66,47 +65,16 @@ pub struct ThreadContext<'a> {
     threads: std::sync::RwLockReadGuard<'a, Vec<std::thread::Thread>>,
     counter: &'a AtomicUsize,
     index: usize,
-
-    /// We operate in 4 phases, depending on the value of `phase % 4`:
-    ///
-    /// `phase` | byte | direction | start | end
-    /// --------|------|-----------|-------|-----
-    ///    0    |   0  |  up       | 0     | `N`
-    ///    1    |   1  |  up       | 0     | `N`
-    ///    2    |   0  |  down     | `N`   | 0
-    ///    3    |   1  |  down     | `N`   | 0
-    ///
-    /// (`N` is `self.threads.len()`)
-    ///
-    /// Note that the pairs of adjacent phases are non-interfering: if a thread
-    /// in phase 0 notices that it has hit `N`, then it can immediately enter
-    /// phase 1 and start modifing byte 1 of the counter, without other threads
-    /// in phase 0 noticing.
-    phase: u8,
 }
 
 impl ThreadContext<'_> {
-    /// Returns `true` if any threads in the pool are sleeping
-    fn any_sleeping(&self) -> bool {
-        let mut v = self.counter.load(Ordering::Acquire);
-        if (self.phase % 2) == 1 {
-            v >>= 8;
-        }
-        // Check to see if any other threads are sleeping; otherwise, skip the
-        // unparking step (because it costs time)
-        (((self.phase % 2) / 2 == 0) && v != 0)
-            || (((self.phase % 2) / 2 == 1) && v != self.threads.len())
-    }
-
     /// If some threads in the pool are sleeping, wakes them up
     ///
     /// This function should be called when work is available.
     pub fn wake(&self) {
-        if self.any_sleeping() {
-            for (i, t) in self.threads.iter().enumerate() {
-                if i != self.index {
-                    t.unpark();
-                }
+        for (i, t) in self.threads.iter().enumerate() {
+            if i != self.index {
+                t.unpark();
             }
         }
     }
@@ -118,9 +86,7 @@ impl ThreadContext<'_> {
     /// (i.e. with `i == self.index`)
     pub fn wake_one(&self, i: usize) {
         assert_ne!(i, self.index);
-        if self.any_sleeping() {
-            self.threads[i].unpark();
-        }
+        self.threads[i].unpark();
     }
 
     /// Used to record that a piece of data has been scheduled for processing
@@ -148,14 +114,8 @@ impl ThreadContext<'_> {
     }
 
     fn done(&self, v: usize) -> bool {
-        (v >> 16 == 0) // No MPSC work queued up
-            & match self.phase % 4 {
-                0 => v & 0xFF == self.threads.len(),
-                1 => v >> 8 == self.threads.len(),
-                2 => v & 0xFF == 0,
-                3 => v >> 8 == 0,
-                _ => unreachable!(),
-            }
+        v >> 16 == 0 // No MPSC work queued up
+            && v >> 8 == self.threads.len() // all threads sleeping
     }
 
     /// Sends the given thread to sleep
@@ -164,13 +124,7 @@ impl ThreadContext<'_> {
     /// threads in the pool have requested to sleep, indicating that all work is
     /// done and they should now halt.
     pub fn sleep(&mut self) -> bool {
-        let v = match self.phase % 4 {
-            0 => self.counter.fetch_add(1, Ordering::Release) + 1,
-            1 => self.counter.fetch_add(256, Ordering::Release) + 256,
-            2 => self.counter.fetch_sub(1, Ordering::Release) - 1,
-            3 => self.counter.fetch_sub(256, Ordering::Release) - 256,
-            _ => unreachable!(),
-        };
+        let v = self.counter.fetch_add(256, Ordering::Release) + 256;
 
         // At this point, the thread doesn't have any work to do, so we'll
         // consider putting it to sleep.  However, if every other thread is
@@ -197,18 +151,11 @@ impl ThreadContext<'_> {
         }
 
         if done {
-            self.phase += 1;
+            false // stop looping
         } else {
-            // Back to the grind
-            match self.phase % 4 {
-                0 => self.counter.fetch_sub(1, Ordering::Release),
-                1 => self.counter.fetch_sub(256, Ordering::Release),
-                2 => self.counter.fetch_add(1, Ordering::Release),
-                3 => self.counter.fetch_add(256, Ordering::Release),
-                _ => unreachable!(),
-            };
+            self.counter.fetch_sub(256, Ordering::Release);
+            true // keep going
         }
-        !done
     }
 }
 
@@ -355,14 +302,12 @@ mod test {
                 s.spawn(move || {
                     let mut ctx = pool.start(i);
                     let t = std::time::Duration::from_millis(1);
-                    for _ in 0..8 {
-                        for _ in 0..i {
-                            std::thread::sleep(t);
-                            ctx.wake();
-                        }
-                        while ctx.sleep() {
-                            // Loop forever
-                        }
+                    for _ in 0..i {
+                        std::thread::sleep(t);
+                        ctx.wake();
+                    }
+                    while ctx.sleep() {
+                        // Loop forever
                     }
                     done.fetch_add(1, Ordering::Release);
                 });
@@ -370,4 +315,48 @@ mod test {
         });
         assert_eq!(done.load(Ordering::Acquire), N);
     }
+
+    #[test]
+    fn queue_and_thread_pool() {
+        const N: usize = 8;
+        let mut queues = QueuePool::new(N);
+        let pool = &ThreadPool::new(N);
+        let mut counters = [0i32; N];
+        const DEPTH: usize = 16;
+        queues[0].push(DEPTH);
+
+        // Confirm that stealing leads to shared work between two threads
+        std::thread::scope(|s| {
+            for (i, (q, c)) in
+                queues.iter_mut().zip(counters.iter_mut()).enumerate()
+            {
+                s.spawn(move || {
+                    let mut ctx = pool.start(i);
+                    loop {
+                        if let Some(v) = q.pop() {
+                            *c += 1;
+                            if v != 0 {
+                                q.push(v - 1);
+                                q.push(v - 1);
+                            }
+                            if q.changed() {
+                                ctx.wake();
+                            }
+                            continue;
+                        }
+                        if !ctx.sleep() {
+                            break;
+                        }
+                    }
+                });
+            }
+        });
+
+        const EXPECTED_COUNT: usize = (1 << (DEPTH + 1)) - 1;
+        assert_eq!(
+            counters.iter().sum::<i32>(),
+            EXPECTED_COUNT as i32,
+            "threads did not complete all work"
+        );
+    }
 }