I have the same problem with every profiled functions.
Profiling time of a function seems to be add to the caller function time :

<?php
function OneThousandMd5() {
for($i=0; $i<1000;$i++) {
$test = md5('dummy');
}
}
OneThousandMd5();
?>

Results :
onThousandMd5 (self) : 36 527 (69,72%)
md5 : 16, 191 (30,28%)

Another example more signifiant :
<?php

/
Only one profiled function within.
/
function oneDoThousandArray() {
doThousandArray();
}

function doThousandArray() {
for($i=0; $i<1000;$i++) {
$test = array();
}
}
/
1000 profiled functions within.
/
function oneThoushandDoArray() {
for($i=0; $i<1000;$i++) {
doArray();
}
}

function doArray() {
$test = array();
}

$s = microtime(true);
oneDoThousandArray();
$e = microtime(true);
echo "oneDoThousandArray : ".(($e - $s) 1000000)."\n";
$s = microtime(true);
oneThoushandDoArray();
$e = microtime(true);
echo "oneThoushandDoArray : ".(($e - $s) 1000000)."\n";

?>

Results :
Without xdebug profiling :
oneDoThousandArray : 965.118408203
oneThoushandDoArray : 14851.0932922

With xdebug profiling
oneDoThousandArray : 1019.9546814 (1.05x)
oneThoushandDoArray : 46550.0354767 (3.13x)

Cachegrind data :
oneDoThousandArray : 100% (0.867 ms)
oneDoThousandArray (self) : 6.81% (1 call)
doThousandArray : 93.19 %
oneThousandDoArray : 100% (47, 374 ms)
oneThousandDoArray (self) : 77.74% (1 call)
doArray : 22.26% (1000 calls)

The best would be to remove profiling time from xdebug cachegrind file, or, if not possible, to add the profiling time of the function to the function executing time not to the caller executing time.