Add note about performance vs correctness

lib/mixins/property-effects.html

@@ -2629,6 +2629,17 @@
        * - Inline computed method (supports negation):
        *   `[[compute(a, 'literal', b)]]`, `[[!compute(a, 'literal', b)]]`
        *
+       * The default implementation uses a regular expression for best
+       * performance. However, the regular expression uses a white-list of
+       * allowed characters in a data-binding, which causes problems for
+       * data-bindings that do use characters not in this white-list.
+       *
+       * Instead of updating the white-list with all allowed characters,
+       * there is a StrictBindingParser (see lib/mixins/strict-binding-parser)
+       * that uses a state machine instead. This state machine is able to handle
+       * all characters. However, it is slightly less performant, therefore we
+       * extracted it into a separate optional mixin.
+       *
        * @param {string} text Text to parse from attribute or textContent
        * @param {Object} templateInfo Current template metadata
        * @return {Array<!BindingPart>} Array of binding part metadata