How permissive is the Windows autorun.inf parsing?

While reading the F-Secure blogpost titled When is AUTORUN.INF really an AUTORUN.INF?, I was reminded of this masking technique - putting extra data between the relevant lines.

But how tolerant is the autorun.inf parser (which I suppose in fact is the INI file parser) really? The example showed by F-Secure is quite mild, in the sense that each line is preceeded by ";" to clearly mark it as a comment. We can up the ante by just putting arbitrary garbage, without the ";" marker:

for (1..255) { print chr(rand(256)); }
print "\n[autorun]\n";
for (1..255) { print chr(rand(256)); }
print "\nopen=C:\\WINDOWS\\NOTEPAD.EXE\n";
for (1..255) { print chr(rand(256)); }

This works! I didn't how large the binary blocks can be, but I imagine that you can go 32k-64k easily. Now to put an extra twist in there:

print "\xFF\xFE";
for (1..255) { print chr(rand(256)); }
print "\n[autorun]\n";
for (1..255) { print chr(rand(256)); }
print "\nopen=C:\\WINDOWS\\NOTEPAD.EXE\n";
for (1..255) { print chr(rand(256)); }

That FF FE is the Byte order mark(er) and tells text editors which recognize it that the file is UTF-16 little endian encoded, which of course it isn't. The end result? You will see only garbage in Notepad (as it tries to "decode" the file) and also in my favorite file management app. And yet, it still works!

Have fun, stay safe and make sure that to disable your autorun.

Update: the commenter had a good idea. In response: strings extracts the strings correctly (I tried it under Linux, but I assume that the Windows version would behave the same). However, for example gedit can't open it.