diff --git a/src/uu/sort/benches/sort_locale_utf8_bench.rs b/src/uu/sort/benches/sort_locale_utf8_bench.rs
index 6f61dc322d0..ea0b0cb9546 100644
--- a/src/uu/sort/benches/sort_locale_utf8_bench.rs
+++ b/src/uu/sort/benches/sort_locale_utf8_bench.rs
@@ -92,6 +92,68 @@ fn sort_unique_utf8_locale(bencher: Bencher) {
     });
 }
 
+/// Benchmark sorting very long lines (single repeated character per line) with UTF-8 locale.
+/// This reproduces the pathological case from issue #12138 where computing full collation
+/// sort keys for multi-megabyte lines caused a 40x slowdown vs GNU sort.
+/// We use 1 MB lines (26 lines, one per letter) to keep the benchmark fast while still
+/// exercising the prefix-based sort key optimization.
+#[divan::bench]
+fn sort_very_long_lines_utf8_locale(bencher: Bencher) {
+    let mut data = Vec::new();
+    // Create 26 lines of 1 MB each, each line is a single repeated letter
+    let letters: Vec<u8> = (b'a'..=b'z').collect();
+    for &ch in &letters {
+        data.extend(std::iter::repeat_n(ch, 1_000_000));
+        data.push(b'\n');
+    }
+    let file_path = setup_test_file(&data);
+    let output_file = NamedTempFile::new().unwrap();
+    let output_path = output_file.path().to_str().unwrap().to_string();
+
+    let args = [
+        "--parallel",
+        "1",
+        "-o",
+        &output_path,
+        file_path.to_str().unwrap(),
+    ];
+    // Warm up
+    black_box(run_util_function(uumain, &args));
+    bencher.bench(|| {
+        black_box(run_util_function(uumain, &args));
+    });
+}
+
+/// Benchmark sorting lines that share a long common prefix but differ after 8 KB,
+/// exercising the fallback from prefix sort keys to full locale comparison.
+#[divan::bench]
+fn sort_long_common_prefix_utf8_locale(bencher: Bencher) {
+    let mut data = Vec::new();
+    let prefix_len = 16 * 1024; // 16 KB common prefix (exceeds the 8 KB sort key limit)
+    let prefix: Vec<u8> = std::iter::repeat_n(b'x', prefix_len).collect();
+    // 26 lines that share the prefix but differ in the suffix
+    for ch in b'a'..=b'z' {
+        data.extend_from_slice(&prefix);
+        data.extend(std::iter::repeat_n(ch, 100));
+        data.push(b'\n');
+    }
+    let file_path = setup_test_file(&data);
+    let output_file = NamedTempFile::new().unwrap();
+    let output_path = output_file.path().to_str().unwrap().to_string();
+
+    let args = [
+        "--parallel",
+        "1",
+        "-o",
+        &output_path,
+        file_path.to_str().unwrap(),
+    ];
+    black_box(run_util_function(uumain, &args));
+    bencher.bench(|| {
+        black_box(run_util_function(uumain, &args));
+    });
+}
+
 fn main() {
     // Set UTF-8 locale BEFORE any benchmarks run.
     // This must happen before divan::main() because the locale is cached