diff --git a/node/rpcstack_bench_test.go b/node/rpcstack_bench_test.go
new file mode 100644
index 0000000000..24bba4a500
--- /dev/null
+++ b/node/rpcstack_bench_test.go
@@ -0,0 +1,210 @@
+package node
+
+import (
+	"fmt"
+	"io"
+	"math"
+	"net/http"
+	"sort"
+	"strings"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/ethereum/go-ethereum/log"
+	"github.com/ethereum/go-ethereum/rpc"
+)
+
+// createAndStartServerTB is will create a server instance with the given config and returns it.
+func createAndStartServerTB(tb testing.TB, httpConfig *httpConfig, disableHTTP2 bool, wsConfig *wsConfig, apis []rpc.API) *httpServer {
+	tb.Helper()
+	srv := &httpServer{
+		log:          log.New("rpc-bench"),
+		timeouts:     rpc.DefaultHTTPTimeouts,
+		disableHTTP2: disableHTTP2,
+	}
+	if err := srv.setListenAddr("127.0.0.1", 0); err != nil {
+		tb.Fatal("failed to set listen address:", err)
+	}
+	if err := srv.enableRPC(apis, *httpConfig); err != nil {
+		tb.Fatal("failed to enable RPC:", err)
+	}
+	if err := srv.enableWS(apis, *wsConfig); err != nil {
+		tb.Fatal("failed to enable WS:", err)
+	}
+	if err := srv.start(); err != nil {
+		tb.Fatal("failed to start HTTP server:", err)
+	}
+	return srv
+}
+
+// newHTTP1Client returns a vanilla HTTP/1.1 client.
+func newHTTP1Client() *http.Client {
+	tr := &http.Transport{
+		MaxIdleConns:        2048,
+		MaxIdleConnsPerHost: 2048,
+		IdleConnTimeout:     90 * time.Second,
+	}
+	tr.Protocols = new(http.Protocols)
+	tr.Protocols.SetHTTP1(true)
+	return &http.Client{Transport: tr}
+}
+
+// newH2CClient returns a client that speaks unencrypted HTTP/2 (H2C).
+func newH2CClient() *http.Client {
+	tr := &http.Transport{}
+	tr.Protocols = new(http.Protocols)
+	tr.Protocols.SetUnencryptedHTTP2(true)
+	return &http.Client{Transport: tr}
+}
+
+// rpcPayload is a minimal valid JSON-RPC 2.0 request.
+const rpcPayload = `{"jsonrpc":"2.0","id":1,"method":"rpc_modules","params":[]}`
+
+// doRequest sends one JSON-RPC POST and discards the body. It returns the
+// round-trip latency in nanoseconds and any error.
+func doRequest(client *http.Client, url string) (int64, error) {
+	start := time.Now()
+	resp, err := client.Post(url, "application/json",
+		strings.NewReader(rpcPayload))
+	if err != nil {
+		return 0, err
+	}
+	defer resp.Body.Close()
+	_, _ = io.Copy(io.Discard, resp.Body) // drain so the connection is reusable
+	if resp.StatusCode != http.StatusOK {
+		return 0, fmt.Errorf("unexpected status %d", resp.StatusCode)
+	}
+	return time.Since(start).Nanoseconds(), nil
+}
+
+// latencyHistogram is a lock-free accumulator for percentile reporting.
+type latencyHistogram struct {
+	mu      sync.Mutex
+	samples []int64
+}
+
+func (h *latencyHistogram) record(ns int64) {
+	h.mu.Lock()
+	h.samples = append(h.samples, ns)
+	h.mu.Unlock()
+}
+
+func (h *latencyHistogram) report(b *testing.B) {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+	if len(h.samples) == 0 {
+		return
+	}
+	sort.Slice(h.samples, func(i, j int) bool { return h.samples[i] < h.samples[j] })
+	n := len(h.samples)
+	p := func(pct float64) float64 {
+		idx := int(math.Ceil(pct/100.0*float64(n))) - 1
+		if idx < 0 {
+			idx = 0
+		}
+		if idx >= n {
+			idx = n - 1
+		}
+		return float64(h.samples[idx]) / 1e6 // ns → ms
+	}
+	b.ReportMetric(p(50), "p50_ms")
+	b.ReportMetric(p(95), "p95_ms")
+	b.ReportMetric(p(99), "p99_ms")
+}
+
+// benchConcurrent is the shared driver used by every sub-benchmark.
+//
+//	client      – pre-configured HTTP client (HTTP/1 or H2C)
+//	url         – full RPC endpoint URL
+//	concurrency – number of parallel goroutines hammering the server
+func benchConcurrent(b *testing.B, client *http.Client, url string, concurrency int) {
+	b.Helper()
+
+	hist := &latencyHistogram{}
+	var errors atomic.Int64
+
+	b.SetParallelism(concurrency)
+	b.ResetTimer()
+
+	b.RunParallel(func(pb *testing.PB) {
+		for pb.Next() {
+			ns, err := doRequest(client, url)
+			if err != nil {
+				errors.Add(1)
+				continue
+			}
+			hist.record(ns)
+		}
+	})
+
+	b.StopTimer()
+	hist.report(b)
+
+	if e := errors.Load(); e > 0 {
+		b.Logf("WARN: %d requests failed", e)
+	}
+}
+
+var concurrencyLevels = []int{1, 10, 50, 100, 500}
+
+// BenchmarkRPCHTTP1 measures the HTTP/1.1 baseline.
+func BenchmarkRPCHTTP1(b *testing.B) {
+	srv := createAndStartServerTB(b, &httpConfig{}, true /*disableHTTP2*/, &wsConfig{}, nil)
+	defer srv.stop()
+
+	url := "http://" + srv.listenAddr()
+	client := newHTTP1Client()
+
+	for _, c := range concurrencyLevels {
+		c := c
+		b.Run(fmt.Sprintf("concurrency=%d", c), func(b *testing.B) {
+			benchConcurrent(b, client, url, c)
+		})
+	}
+}
+
+// BenchmarkRPCH2C measures HTTP/2 cleartext (H2C).
+func BenchmarkRPCH2C(b *testing.B) {
+	srv := createAndStartServerTB(b, &httpConfig{}, false /*disableHTTP2*/, &wsConfig{}, nil)
+	defer srv.stop()
+
+	url := "http://" + srv.listenAddr()
+	client := newH2CClient()
+
+	for _, c := range concurrencyLevels {
+		c := c
+		b.Run(fmt.Sprintf("concurrency=%d", c), func(b *testing.B) {
+			benchConcurrent(b, client, url, c)
+		})
+	}
+}
+
+// BenchmarkRPCH2CvsHTTP1SameServer exercises both protocols against the same
+// server instance (H2C enabled) to isolate protocol overhead from any
+// server-startup variance.
+func BenchmarkRPCH2CvsHTTP1SameServer(b *testing.B) {
+	srv := createAndStartServerTB(b, &httpConfig{}, false /*disableHTTP2*/, &wsConfig{}, nil)
+	defer srv.stop()
+
+	url := "http://" + srv.listenAddr()
+
+	protocols := []struct {
+		name   string
+		client *http.Client
+	}{
+		{"HTTP1", newHTTP1Client()},
+		{"H2C", newH2CClient()},
+	}
+
+	for _, proto := range protocols {
+		proto := proto
+		for _, c := range concurrencyLevels {
+			c := c
+			b.Run(fmt.Sprintf("%s/concurrency=%d", proto.name, c), func(b *testing.B) {
+				benchConcurrent(b, proto.client, url, c)
+			})
+		}
+	}
+}