diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 5cd5243e7b..de50bec4a5 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -244,14 +244,23 @@ jobs:
 
     - name: Run tests under Miri
       run: |
+        set -eo pipefail
+
         # Work around https://github.com/rust-lang/miri/issues/3125
         [ "${{ matrix.target }}" == "aarch64-unknown-linux-gnu" ] && cargo clean
-      
+
+        # Spawn twice the number of workers as there are CPU cores.
+        THREADS=$(echo "$(nproc) * 2" | bc)
+        echo "Running Miri tests with $THREADS threads" | tee -a $GITHUB_STEP_SUMMARY
+
+        cargo install cargo-nextest
+
         # Run under both the stacked borrows model (default) and under the tree 
         # borrows model to ensure we're compliant with both.
         for EXTRA_FLAGS in "" "-Zmiri-tree-borrows"; do
           MIRIFLAGS="$MIRIFLAGS $EXTRA_FLAGS" ./cargo.sh +${{ matrix.toolchain }} \
-            miri test \
+            miri nextest run \
+            --test-threads "$THREADS" \
             --package ${{ matrix.crate }} \
             --target ${{ matrix.target }} \
             ${{ matrix.features }}
@@ -464,6 +473,7 @@ jobs:
           cargo metadata                             &> /dev/null &
           cargo install cargo-readme --version 3.2.0 &> /dev/null &
           cargo install --locked kani-verifier       &> /dev/null &
+          cargo install cargo-nextest                &> /dev/null &
           cargo kani setup                           &> /dev/null &
 
           wait
diff --git a/src/pointer/ptr.rs b/src/pointer/ptr.rs
index 523cdfe34d..7459b79371 100644
--- a/src/pointer/ptr.rs
+++ b/src/pointer/ptr.rs
@@ -1526,8 +1526,9 @@ mod tests {
         }
     }
 
-    #[test]
-    fn test_ptr_try_cast_into_soundness() {
+    mod test_ptr_try_cast_into_soundness {
+        use super::*;
+
         // This test is designed so that if `Ptr::try_cast_into_xxx` are
         // buggy, it will manifest as unsoundness that Miri can detect.
 
@@ -1650,9 +1651,21 @@ mod tests {
             trailing: [T],
         }
 
+        // Each test case becomes its own `#[test]` function. We do this because
+        // this test in particular takes far, far longer to execute under Miri
+        // than all of our other tests combined. Previously, we had these
+        // execute sequentially in a single test function. We run Miri tests in
+        // parallel in CI, but this test being sequential meant that most of
+        // that parallelism was wasted, as all other tests would finish in a
+        // fraction of the total execution time, leaving this test to execute on
+        // a single thread for the remainder of the test. By putting each test
+        // case in its own function, we permit better use of available
+        // parallelism.
         macro_rules! test {
-            ($($ty:ty),*) => {
-                $({
+            ($test_name:ident: $ty:ty) => {
+                #[test]
+                #[allow(non_snake_case)]
+                fn $test_name() {
                     const S: usize = core::mem::size_of::<$ty>();
                     const N: usize = if S == 0 { 4 } else { S * 4 };
                     test::<$ty, _, N>([None]);
@@ -1667,11 +1680,15 @@ mod tests {
                         test::<[$ty], _, N>([None, Some(0), Some(1), Some(2), Some(3)]);
                         test::<SliceDst<$ty>, _, N>([None, Some(0), Some(1), Some(2), Some(3)]);
                     }
-                })*
+                }
+            };
+            ($ty:ident) => {
+                test!($ty: $ty);
             };
+            ($($ty:ident),*) => { $(test!($ty);)* }
         }
 
-        test!(());
+        test!(empty_tuple: ());
         test!(u8, u16, u32, u64, u128, usize, AU64);
         test!(i8, i16, i32, i64, i128, isize);
         test!(f32, f64);