diff --git a/performance.md b/performance.md
new file mode 100644
--- /dev/null
+++ b/performance.md
@@ -0,0 +1,27 @@
+# Performance comparison #
+## The base case ##
+
+Let's assume we have a secret of length _m_. The splitting takes _n_ evaluations of a polynomial of order _k_ (over Galois field 256) for each byte, leading to _O(n\*k\*m)_ finite field multiplications. Reconstruction of the constant parameters during joining first precomputes parts of the Lagrange polynomial and then reuses them for each byte, taking _O(k\*k + k\*m)_ multiplications.
+
+Benchmark results. The times for split and join mean _seconds per byte_ of the secret length:
+<table>
+    <tr>
+        <th>Revision</th>
+        <th>Features</th>
+        <th>k / n parameters</th>
+        <th>Split</th>
+        <th>Join</th>
+    </tr>
+    <tr>
+        <td rowspan="2">a47ae3e113cc</td>
+        <td rowspan="2">-</td>
+        <td>2 / 3</td>
+        <td>5.02e-06</td>
+        <td>4.12e-05</td>
+    </tr>
+    <tr>
+        <td>254 / 254</td>
+        <td>0.0125</td>
+        <td>0.00175</td>
+    </tr>
+</table>
\ No newline at end of file
diff --git a/readme.md b/readme.md
--- a/readme.md
+++ b/readme.md
@@ -26,8 +26,8 @@ Benchmark results, all values mean _seco
     </tr>
     <tr>
         <td>2 / 3 (a laptop)</td>
-        <td>8.7e-06</td>
-        <td>7.17e-05</td>
+        <td>5.02e-06</td>
+        <td>4.12e-05</td>
     </tr>
     <tr>
         <td>254 / 254 (a Raspberry Pi 3)</td>
@@ -36,8 +36,8 @@ Benchmark results, all values mean _seco
     </tr>
     <tr>
         <td>254 / 254 (a laptop)</td>
-        <td>0.0209</td>
-        <td>0.00347</td>
+        <td>0.0125</td>
+        <td>0.00175</td>
     </tr>
 </table>