diff --git a/include/sys/arc_impl.h b/include/sys/arc_impl.h
index 6134ac1f5..dbe712e2e 100644
--- a/include/sys/arc_impl.h
+++ b/include/sys/arc_impl.h
@@ -469,6 +469,12 @@ typedef struct l2arc_dev {
 	boolean_t		l2ad_thread_exit;	/* signal thread exit */
 	kmutex_t		l2ad_feed_thr_lock;	/* thread sleep/wake */
 	kcondvar_t		l2ad_feed_cv;		/* thread wakeup cv */
+	/*
+	 * Consecutive cycles where metadata filled write budget
+	 * while data passes got nothing written. Used to detect
+	 * monopolization and skip metadata to give data a chance.
+	 */
+	uint64_t		l2ad_meta_cycles;
 } l2arc_dev_t;
 
 /*
diff --git a/man/man4/zfs.4 b/man/man4/zfs.4
index d01a36821..509b3109a 100644
--- a/man/man4/zfs.4
+++ b/man/man4/zfs.4
@@ -233,6 +233,13 @@ resetting its markers to the tail.
 Lower values keep the marker closer to the tail under active workloads.
 Set to 0 to disable the depth cap.
 .
+.It Sy l2arc_meta_cycles Ns = Ns Sy 2 Pq u64
+How many consecutive cycles metadata may monopolize the write budget
+before being skipped to let data run.
+The default of 2 gives metadata roughly 67% and data 33% of L2ARC
+write bandwidth under sustained load.
+Higher values favor metadata; set to 0 to disable.
+.
 .It Sy l2arc_write_max Ns = Ns Sy 33554432 Ns B Po 32 MiB Pc Pq u64
 Maximum write rate in bytes per second for each L2ARC device.
 Used directly during initial fill, when DWPD limiting is disabled,
diff --git a/module/zfs/arc.c b/module/zfs/arc.c
index d59ce0225..166304959 100644
--- a/module/zfs/arc.c
+++ b/module/zfs/arc.c
@@ -963,6 +963,15 @@ static int l2arc_mfuonly = 0;
  */
 static uint64_t l2arc_ext_headroom_pct = 25;
 
+/*
+ * Metadata monopolization limit.  When metadata fills the write budget
+ * for this many consecutive cycles while data gets nothing, skip metadata
+ * for one cycle to let data run, then reset the counter.
+ * With N=2, the steady-state pattern under sustained monopolization is
+ * 2 metadata cycles followed by 1 data cycle (67%/33% split).
+ */
+static uint64_t l2arc_meta_cycles = 2;
+
 /*
  * L2ARC TRIM
  * l2arc_trim_ahead : A ZFS module parameter that controls how much ahead of
@@ -9998,6 +10007,12 @@ l2arc_write_buffers(spa_t *spa, l2arc_dev_t *dev, uint64_t target_sz)
 	/*
 	 * Copy buffers for L2ARC writing.
 	 */
+	boolean_t skip_meta = (save_position &&
+	    l2arc_meta_cycles > 0 &&
+	    dev->l2ad_meta_cycles >= l2arc_meta_cycles);
+	if (skip_meta)
+		dev->l2ad_meta_cycles = 0;
+
 	for (int pass = 0; pass < L2ARC_FEED_TYPES; pass++) {
 		/*
 		 * pass == 0: MFU meta
@@ -10013,6 +10028,9 @@ l2arc_write_buffers(spa_t *spa, l2arc_dev_t *dev, uint64_t target_sz)
 				continue;
 		}
 
+		if (skip_meta && pass <= L2ARC_MRU_META)
+			continue;
+
 		headroom = target_sz * l2arc_headroom;
 		if (zfs_compressed_arc_enabled)
 			headroom = (headroom * l2arc_headroom_boost) / 100;
@@ -10080,6 +10098,14 @@ l2arc_write_buffers(spa_t *spa, l2arc_dev_t *dev, uint64_t target_sz)
 		    (spa->spa_l2arc_info.l2arc_next_sublist[pass] + 1) %
 		    num_sublists;
 
+		/*
+		 * Count consecutive metadata monopolization toward
+		 * l2arc_meta_cycles.  Only count when metadata actually
+		 * filled the write budget, starving data passes.
+		 */
+		if (save_position && pass <= L2ARC_MRU_META && full)
+			dev->l2ad_meta_cycles++;
+
 		/*
 		 * Depth cap: track cumulative bytes scanned per pass
 		 * and reset markers when the scan cap is reached.
@@ -10109,6 +10135,13 @@ l2arc_write_buffers(spa_t *spa, l2arc_dev_t *dev, uint64_t target_sz)
 			break;
 	}
 
+	/*
+	 * If nothing was written at all, reset monopolization counter.
+	 * No point skipping metadata if data has nothing either.
+	 */
+	if (write_asize == 0)
+		dev->l2ad_meta_cycles = 0;
+
 	/* No buffers selected for writing? */
 	if (pio == NULL) {
 		ASSERT0(write_psize);
@@ -11752,6 +11785,9 @@ ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, mfuonly, INT, ZMOD_RW,
 ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, exclude_special, INT, ZMOD_RW,
 	"Exclude dbufs on special vdevs from being cached to L2ARC if set.");
 
+ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, meta_cycles, U64, ZMOD_RW,
+	"Consecutive metadata cycles before skipping to let data run");
+
 ZFS_MODULE_PARAM(zfs_l2arc, l2arc_, ext_headroom_pct, U64, ZMOD_RW,
 	"Depth cap as percentage of state size for marker reset");
 
diff --git a/tests/zfs-tests/include/tunables.cfg b/tests/zfs-tests/include/tunables.cfg
index 4008bfe2b..e2b829b84 100644
--- a/tests/zfs-tests/include/tunables.cfg
+++ b/tests/zfs-tests/include/tunables.cfg
@@ -53,6 +53,7 @@ L2ARC_NOPREFETCH		l2arc.noprefetch		l2arc_noprefetch
 L2ARC_REBUILD_BLOCKS_MIN_L2SIZE	l2arc.rebuild_blocks_min_l2size	l2arc_rebuild_blocks_min_l2size
 L2ARC_REBUILD_ENABLED		l2arc.rebuild_enabled		l2arc_rebuild_enabled
 L2ARC_TRIM_AHEAD		l2arc.trim_ahead		l2arc_trim_ahead
+L2ARC_META_CYCLES		l2arc.meta_cycles		l2arc_meta_cycles
 L2ARC_WRITE_MAX			l2arc.write_max			l2arc_write_max
 LIVELIST_CONDENSE_NEW_ALLOC	livelist.condense.new_alloc	zfs_livelist_condense_new_alloc
 LIVELIST_CONDENSE_SYNC_CANCEL	livelist.condense.sync_cancel	zfs_livelist_condense_sync_cancel