Your problem is the .*? -- although you rightly tried to limit the number of don't care characters using '?' your match still grabs the smallest number of don't care characters between '<!--' and 'DOS'... which just happens to include another '-->' and '<!--', the pieces it isn't supposed to grab.

You'll need a few baselines to come up with a workable regexp:

• Can xxxx ever include '<!--' or '-->'? (We'd better hope not)
• Can xxxx contain only word-like characters (\w)?
• Or can it include space characters as well (\s)?

These would help you optimize your regexp...

But the key point is that you need to limit your regexp to a single comment group containing 'DOS'.

I'd suggest using:

$html = '<!--% xxxx UNO xxxx %-->   <!--% xxxx DOS xxxx %-->';
$html =~ s/<!--%([^->]+?) DOS ([^->]+)%-->/GONE/s;
print $html;
[download]

It's a little ugly, and notice that it assumes that your 'xxxx' can't contain '->', which may or may not be the case.

YMMV