diff --git a/core/state/trie_prefetcher_test.go b/core/state/trie_prefetcher_test.go
index 655ffc3544..7650bb2e7d 100644
--- a/core/state/trie_prefetcher_test.go
+++ b/core/state/trie_prefetcher_test.go
@@ -19,6 +19,7 @@ package state
 import (
 	"math/big"
 	"sync"
+	"sync/atomic"
 	"testing"
 
 	"github.com/ethereum/go-ethereum/common"
@@ -71,8 +72,8 @@ func TestSchdeulerTerminationRaceCondition(t *testing.T) {
 	// The lock-based implementation of [subfetcher] had a race condition
 	// whereby schedule() could obtain the lock after the <-sf.stop branch of
 	// loop() had already checked for an empty queue. Although probabilistic,
-	// this test reliably triggered at a rate of ~4 in 10,000 on an Apple M3 Max
-	// chip.
+	// this test reliably triggered at a rates of (~4 in 10k) and (~100 in 50k)
+	// on an Apple M3 Max chip.
 
 	t.Parallel()
 	db := filledStateDB()
@@ -80,9 +81,13 @@ func TestSchdeulerTerminationRaceCondition(t *testing.T) {
 
 	// Maximise concurrency by synchronising all scheduling and termination.
 	start := make(chan struct{})
-	var wg sync.WaitGroup
+	var (
+		wg          sync.WaitGroup
+		raceInduced atomic.Uint64
+	)
 
-	for i := 0; i < 50_000; i++ {
+	const numTrials = 50_000
+	for i := 0; i < numTrials; i++ {
 		wg.Add(2)
 		fetcher := newSubfetcher(db.db, db.originalRoot, common.Hash{}, db.originalRoot, common.Address{})
 
@@ -102,13 +107,16 @@ func TestSchdeulerTerminationRaceCondition(t *testing.T) {
 
 			<-doneScheduling
 			if gotScheduleErr == nil && len(fetcher.tasks) > 0 {
-				t.Errorf("%T.schedule() returned nil error but %d task(s) remain in queue after %T.terminate([blocking]) returned", fetcher, len(fetcher.tasks), fetcher)
+				raceInduced.Add(1)
 			}
 		}()
 	}
 
 	close(start)
 	wg.Wait()
+	if got := raceInduced.Load(); got > 0 {
+		t.Errorf("In %d/%d concurrent trials %T.schedule() returned nil error but >0 tasks remain in queue after %[3]T.terminate([blocking]) returned", got, numTrials, &subfetcher{})
+	}
 }
 
 func TestVerklePrefetcher(t *testing.T) {