diff --git a/pkg/espflasher/protocol.go b/pkg/espflasher/protocol.go
index d3cf58c..dc67350 100644
--- a/pkg/espflasher/protocol.go
+++ b/pkg/espflasher/protocol.go
@@ -76,6 +76,7 @@ const (
 	chipEraseTimeout    = 120 * time.Second
 	md5Timeout          = 30 * time.Second
 	eraseWritePerMBRate = 10 * time.Second // per megabyte
+	flashWritePerMBRate = 40 * time.Second // per megabyte
 )
 
 // conn wraps the serial port and provides the low-level protocol operations.
@@ -424,7 +425,8 @@ func (c *conn) flashData(block []byte, seq uint32) error {
 	binary.LittleEndian.PutUint32(data[12:16], 0)
 	copy(data[16:], block)
 
-	_, err := c.checkCommand("write flash block", cmdFlashData, data, checksum(block), defaultTimeout, 0)
+	timeout := flashWriteTimeoutForSize(uint32(len(block)))
+	_, err := c.checkCommand("write flash block", cmdFlashData, data, checksum(block), timeout, 0)
 	return err
 }
 
@@ -675,13 +677,14 @@ func md5TimeoutForSize(size uint32) time.Duration {
 }
 
 // flashWriteTimeoutForSize calculates an appropriate ack timeout for
-// compressed flash write/finish commands, scaled by data size using the
-// same per-MB rate as eraseTimeoutForSize. Unlike erase (which floors at
-// 10s for a whole-region operation), write/finish acks are per-block or
-// per-image and use the smaller defaultTimeout floor so small writes
-// aren't over-inflated.
+// flash write/finish commands, scaled by data size. Flash writes (and
+// especially compressed writes that must be decompressed before being
+// programmed) can take substantially longer than the erase rate on slower
+// ESP32 processors, so this uses a more generous per-MB rate than
+// eraseTimeoutForSize. The floor remains defaultTimeout so small blocks
+// are not over-inflated.
 func flashWriteTimeoutForSize(size uint32) time.Duration {
-	t := defaultTimeout + time.Duration(float64(eraseWritePerMBRate)*float64(size)/float64(1024*1024))
+	t := defaultTimeout + time.Duration(float64(flashWritePerMBRate)*float64(size)/float64(1024*1024))
 	if t < defaultTimeout {
 		t = defaultTimeout
 	}
diff --git a/pkg/espflasher/protocol_test.go b/pkg/espflasher/protocol_test.go
index 741bda3..89abd76 100644
--- a/pkg/espflasher/protocol_test.go
+++ b/pkg/espflasher/protocol_test.go
@@ -110,9 +110,10 @@ func TestEraseTimeoutForSize(t *testing.T) {
 }
 
 func TestFlashWriteTimeoutForSize(t *testing.T) {
-	// E1-15 regression: compressed flash write/finish acks must scale by
-	// size (mirroring eraseTimeoutForSize), with a floor of defaultTimeout
-	// so small blocks aren't over-inflated.
+	// E1-15 regression: flash write/finish acks must scale by size, with a
+	// floor of defaultTimeout so small blocks aren't over-inflated. Flash
+	// writes use a more generous per-MB rate than erase because the stub
+	// must decompress and program each block before acking.
 	tests := []struct {
 		name string
 		size uint32
@@ -131,9 +132,9 @@ func TestFlashWriteTimeoutForSize(t *testing.T) {
 		})
 	}
 
-	// A 4MB compressed write must get a materially larger timeout than a
-	// 4KB block, so large transfers over slow USB-serial bridges don't
-	// time out on a flat ack wait.
+	// A 4MB write must get a materially larger timeout than a 4KB block,
+	// so large transfers over slow USB-serial bridges don't time out on a
+	// flat ack wait.
 	small := flashWriteTimeoutForSize(4 * 1024)
 	large := flashWriteTimeoutForSize(4 * 1024 * 1024)
 	if large <= small {
@@ -142,6 +143,14 @@ func TestFlashWriteTimeoutForSize(t *testing.T) {
 	if large < 2*small {
 		t.Errorf("expected 4MB timeout to be materially larger than 4KB timeout: small=%v large=%v", small, large)
 	}
+
+	// Flash writes are given a longer per-MB budget than erase to account
+	// for decompression + programming on slower ESP32 processors.
+	erase := eraseTimeoutForSize(1024 * 1024)
+	write := flashWriteTimeoutForSize(1024 * 1024)
+	if write <= erase {
+		t.Errorf("expected flash write timeout > erase timeout for same size: erase=%v write=%v", erase, write)
+	}
 }
 
 func TestSendCommandFormat(t *testing.T) {