Add speed-accuracy trade-off info to readme

README.md

@@ -32,4 +32,9 @@ use approx::assert_abs_diff_eq;
 let w = lambert_w_0(PI)?;
 
 assert_abs_diff_eq!(w, 1.0736581947961492, epsilon = 1e-7);
-```
+```
+
+## Speed-accuracy trade-off
+
+The 50-bit accurate versions in the `accurate` module are more accurate, but slightly slower, than the 24-bit accurate versions in the `fast` module.
+`fast::lambert_w_0` is around 15% faster than `accurate::lambert_w_0` and `fast::lambert_w_m1` is around 41% faster than `accurate::lambert_w_m1`.