diff --git a/core/txpool/blobpool/blobpool.go b/core/txpool/blobpool/blobpool.go
index 5cf1218b75..4a0296dab9 100644
--- a/core/txpool/blobpool/blobpool.go
+++ b/core/txpool/blobpool/blobpool.go
@@ -55,6 +55,11 @@ const (
 	// tiny overflows causing all txs to move a shelf higher, wasting disk space.
 	txAvgSize = 4 * 1024
 
+	// txBlobOverhead is an approximation of the overhead that an additional blob
+	// has on transaction size. This is added to the slotter to avoid
+	// tiny overflows causing all txs to move a shelf higher, wasting disk space.
+	txBlobOverhead = 2048
+
 	// txMaxSize is the maximum size a single transaction can have, outside
 	// the included blobs. Since blob transactions are pulled instead of pushed,
 	// and only a small metadata is kept in ram, the rest is on disk, there is
diff --git a/core/txpool/blobpool/slotter.go b/core/txpool/blobpool/slotter.go
index 84ccc0f27b..822584cfbd 100644
--- a/core/txpool/blobpool/slotter.go
+++ b/core/txpool/blobpool/slotter.go
@@ -27,10 +27,10 @@ package blobpool
 // of resources, but it makes stress testing with junk transactions simpler.
 func newSlotter(maxBlobsPerTransaction int) func() (uint32, bool) {
 	slotsize := uint32(txAvgSize)
-	slotsize -= uint32(blobSize) // underflows, it's ok, will overflow back in the first return
+	slotsize -= uint32(blobSize) + txBlobOverhead // underflows, it's ok, will overflow back in the first return
 
 	return func() (size uint32, done bool) {
-		slotsize += blobSize
+		slotsize += blobSize + txBlobOverhead
 		finished := slotsize > uint32(maxBlobsPerTransaction)*blobSize+txMaxSize
 
 		return slotsize, finished
diff --git a/core/txpool/blobpool/slotter_test.go b/core/txpool/blobpool/slotter_test.go
index 8d46f47d2c..4e0177c732 100644
--- a/core/txpool/blobpool/slotter_test.go
+++ b/core/txpool/blobpool/slotter_test.go
@@ -33,21 +33,21 @@ func TestNewSlotter(t *testing.T) {
 	}
 	// Compare the database shelves to the expected ones
 	want := []uint32{
-		0*blobSize + txAvgSize,  // 0 blob + some expected tx infos
-		1*blobSize + txAvgSize,  // 1 blob + some expected tx infos
-		2*blobSize + txAvgSize,  // 2 blob + some expected tx infos (could be fewer blobs and more tx data)
-		3*blobSize + txAvgSize,  // 3 blob + some expected tx infos (could be fewer blobs and more tx data)
-		4*blobSize + txAvgSize,  // 4 blob + some expected tx infos (could be fewer blobs and more tx data)
-		5*blobSize + txAvgSize,  // 1-6 blobs + unexpectedly large tx infos < 4 blobs + max tx metadata size
-		6*blobSize + txAvgSize,  // 1-6 blobs + unexpectedly large tx infos < 4 blobs + max tx metadata size
-		7*blobSize + txAvgSize,  // 1-6 blobs + unexpectedly large tx infos < 4 blobs + max tx metadata size
-		8*blobSize + txAvgSize,  // 1-6 blobs + unexpectedly large tx infos < 4 blobs + max tx metadata size
-		9*blobSize + txAvgSize,  // 1-6 blobs + unexpectedly large tx infos < 4 blobs + max tx metadata size
-		10*blobSize + txAvgSize, // 1-6 blobs + unexpectedly large tx infos < 4 blobs + max tx metadata size
-		11*blobSize + txAvgSize, // 1-6 blobs + unexpectedly large tx infos < 4 blobs + max tx metadata size
-		12*blobSize + txAvgSize, // 1-6 blobs + unexpectedly large tx infos < 4 blobs + max tx metadata size
-		13*blobSize + txAvgSize, // 1-6 blobs + unexpectedly large tx infos < 4 blobs + max tx metadata size
-		14*blobSize + txAvgSize, // 1-6 blobs + unexpectedly large tx infos >= 4 blobs + max tx metadata size
+		0*blobSize + 0*txBlobOverhead + txAvgSize,   // 0 blob + some expected tx infos
+		1*blobSize + 1*txBlobOverhead + txAvgSize,   // 1 blob + some expected tx infos
+		2*blobSize + 2*txBlobOverhead + txAvgSize,   // 2 blob + some expected tx infos (could be fewer blobs and more tx data)
+		3*blobSize + 3*txBlobOverhead + txAvgSize,   // 3 blob + some expected tx infos (could be fewer blobs and more tx data)
+		4*blobSize + 4*txBlobOverhead + txAvgSize,   // 4 blob + some expected tx infos (could be fewer blobs and more tx data)
+		5*blobSize + 5*txBlobOverhead + txAvgSize,   // 1-6 blobs + unexpectedly large tx infos < 4 blobs + max tx metadata size
+		6*blobSize + 6*txBlobOverhead + txAvgSize,   // 1-6 blobs + unexpectedly large tx infos < 4 blobs + max tx metadata size
+		7*blobSize + 7*txBlobOverhead + txAvgSize,   // 1-6 blobs + unexpectedly large tx infos < 4 blobs + max tx metadata size
+		8*blobSize + 8*txBlobOverhead + txAvgSize,   // 1-6 blobs + unexpectedly large tx infos < 4 blobs + max tx metadata size
+		9*blobSize + 9*txBlobOverhead + txAvgSize,   // 1-6 blobs + unexpectedly large tx infos < 4 blobs + max tx metadata size
+		10*blobSize + 10*txBlobOverhead + txAvgSize, // 1-6 blobs + unexpectedly large tx infos < 4 blobs + max tx metadata size
+		11*blobSize + 11*txBlobOverhead + txAvgSize, // 1-6 blobs + unexpectedly large tx infos < 4 blobs + max tx metadata size
+		12*blobSize + 12*txBlobOverhead + txAvgSize, // 1-6 blobs + unexpectedly large tx infos < 4 blobs + max tx metadata size
+		13*blobSize + 13*txBlobOverhead + txAvgSize, // 1-6 blobs + unexpectedly large tx infos < 4 blobs + max tx metadata size
+		14*blobSize + 14*txBlobOverhead + txAvgSize, // 1-6 blobs + unexpectedly large tx infos >= 4 blobs + max tx metadata size
 	}
 	if len(shelves) != len(want) {
 		t.Errorf("shelves count mismatch: have %d, want %d", len(shelves), len(want))