Micro optimization isn't worth it.

You had to do it ten million times to notice a difference, a little more than 2 seconds

$a===NULL; Took: 1.2424390316s
 is_null($a); Took: 3.70693397522s

difference = 2.46449494362
difference/10,000,000 = 0.000000246449494362

The execution time difference between ===NULL and is_null is less than 250 nanoseconds. Go optimize something that matters.

See how php parses different values. $var is the variable.

$var        =    NULL    ""    0    "0"    1

strlen($var)    =    0    0    1    1    1
is_null($var)    =    TRUE    FALSE    FALSE    FALSE    FALSE
$var == ""    =    TRUE    TRUE    TRUE    FALSE    FALSE
!$var        =    TRUE    TRUE    TRUE    TRUE    FALSE
!is_null($var)    =    FALSE    TRUE    TRUE    TRUE    TRUE
$var != ""    =    FALSE    FALSE    FALSE    TRUE    TRUE
$var        =    FALSE    FALSE    FALSE    FALSE    TRUE

Peace!

In PHP 7 (phpng), is_null is actually marginally faster than ===, although the performance difference between the two is far smaller.

PHP 5.5.9
is_null - float(2.2381200790405)
===     - float(1.0024659633636)
=== faster by ~100ns per call

PHP 7.0.0-dev (built: May 19 2015 10:16:06)
is_null - float(1.4121870994568)
===     - float(1.4577329158783)
is_null faster by ~5ns per call

A quick test in 2022 on PHP 8.1 confirms there is still no need to micro-optimize NULL checks:

<?php

// Comparison Operator
$before = microtime(true);
$var = null;
for ($i=0 ; $i<1000000000 ; $i++) {
    if($var === null) {}
}
$after = microtime(true);
echo '    ===: ' . ($after - $before) . " seconds\n";

// Function
$before = microtime(true);
$var = null;
for ($i=0 ; $i<1000000000 ; $i++) {
    if(is_null($var)) {}
}
$after = microtime(true);
echo 'is_null: ' . ($after - $before) . " seconds\n";

//     ===: 4.1487579345703 seconds
// is_null: 4.1316878795624 seconds

For what I realized is that  is_null($var)  returns exactly the opposite of  isset($var) , except that is_null($var) throws a notice if $var hasn't been set yet.

the following will prove that:

<?php

$quirks = array(null, true, false, 0, 1, '', "\0", "unset");

foreach($quirks as $var) {
    if ($var === "unset") unset($var);

    echo is_null($var) ? 1 : 0;
    echo isset($var) ? 1 : 0;
    echo "\n";
}

?>

this will print out something like:

10    // null
01    // true
01    // false
01    // 0
01    // 1
01    // ''
01    // "\0"
Notice:  Undefined variable: var in /srv/www/htdocs/sandbox/null/nulltest.php on line 8
10    // (unset)

For the major quirky types/values is_null($var) obviously always returns the opposite of isset($var), and the notice clearly points out the faulty line with the is_null() statement. You might want to examine the return value of those functions in detail, but since both are specified to return boolean types there should be no doubt.

A second look into the PHP specs tells that is_null() checks whether a value is null or not. So, you may pass any VALUE to it, eg. the result of a function.
isset() on the other hand is supposed to check for a VARIABLE's existence, which makes it a language construct rather than a function. Its sole porpuse lies in that checking. Passing anything else will result in an error.

Knowing that, allows us to draw the following unlikely conclusion:

isset() as a language construct is way faster, more reliable and powerful than is_null() and should be prefered over is_null(), except for when you're directly passing a function's result, which is considered bad programming practice anyways.