brandelf brands binary with EI_OSABI but ld doesn't see it if
	old brand exists on it. This is the case of current acroread4
	port.

	First, untouched binary in tarball is:

$ ls -l acroread
-rwxr-xr-x  1 k5  wheel  3379910 Jan 27 09:30 acroread*
$ hd acroread|head -1
00000000  7f 45 4c 46 01 01 01 00  00 00 00 00 00 00 00 00  |.ELF............|

	This hasn't been branded but works on FreeBSD.

	Next, the binary was stripped during make install:

$ ls -l acroread
-rwxr-xr-x  1 k5  wheel  2549840 May 11 00:47 acroread*
$ hd acroread|head -1
00000000  7f 45 4c 46 01 01 01 09  46 72 65 65 42 53 44 00  |.ELF....FreeBSD.|

	strip(1) brands it to FreeBSD with both new and old method.
	This is not so good.

	Then, brandelf it to Linux at post-install target:

$ hd acroread|head -1
00000000  7f 45 4c 46 01 01 01 03  46 72 65 65 42 53 44 00  |.ELF....FreeBSD.|

	Properly branded to Linux with new method but old brand string
	is still there. The problem occurrs here; ld looks old method
	prior to new one for compatibility reason. So this acroread
	binary would be treated as FreeBSD native and then bombed.

	To avoid this confusion: use linux strip instead.
	Or wait for October when old brand method support would be
	removed. :)

	Also strip(1) should not use old method to brand binary I think.

	I don't know about Alpha.

How-To-Repeat: 
/usr/ports/print/acroread4# make install
/usr/ports/print/acroread4# exit
$ acroread4
ELF interpreter /lib/ld-linux.so.2 not found
Abort trap