Given an array, normalize it to remove recursive arrays and return the number of distinct vowel in the word that has more distinct vowels

max_vowel_counter.js

var merge_recursive = function(a, res) {
	if (!Array.isArray(res)) res = [];
	// Using BFS algorithm as if it was a tree
	for (var i = 0; i < a.length;  i++) {
		if (Array.isArray(a[i])) {
			merge_recursive(a[i], res);
		} else {
			res.push(a[i]);
		}
	}
	return res;
}

var vowel_max_counter = function(arr) {
	var tokens = merge_recursive(arr);
	var count_different_vowels = function(str) {
		var vowel_variations = [/[aáãàâä]/, /[eéêèë]/, /[iíîìï]/, 
					/[oóõòôö]/, /[uúùûü]/];
		
		var different_vowels = 0;

		for (var i = 0; i < vowel_variations.length; i++) {
			if (str.match(vowel_variations[i])) {
				different_vowels++;
			}
		}
		
		return different_vowels;
	};

	var max = 0;
	var different_vowels;
	for (var i = 0; i < tokens.length; i++) {
		max = Math.max(count_different_vowels(tokens[i]), max);
	}

	return max;
}

console.info(vowel_max_counter(["hola", ["soy", ["juan", "fernandez"] ], "y", ["no", "tengo", ["dinero"] ] ]));

En http://jsperf.com/gejs-estadistica-de-vocales/4 podemos apreciar lo inconveniente que resulta utilizar RegExp.prototype.exec si lo único que queremos realizar es una comprobación de matching y no nos interesa ningún detalle de éste. En estos casos sin duda, lo que hay que utilizar es RegExp.prototype.test.