>Is there a good reason to not change to GNU m4(Other than religious ones?)?

There are the standard licensing issues. 

>of autoconf says it maybe makes problems with BSDish m4's.

I haven't looked at the NetBSD m4 myself, but the problem with the
System V m4 (back under Apollo Domain/OS) was that it dumped core on
any macro expansion over 1024 bytes. (malloc? realloc? range checking?
sigh...) so it would certainly fail to work with autoconf. There are a
few syntactic ambiguities handled in gnum4 (instead of giving errors
or null substitutions), and the fact that the autoconf 2.x releases
use the m4 --freeze/restore support, but they could probably be worked
around; a start might be to test autoconf against gnu m4 using "m4
--traditional" first.

If anyone is particularly interested in the exercise, I can probably
offer direct advice. (MIT Kerberos V uses autoconf now, so there is at
least one reason to bother...)
			_Mark_ <eichin@kitten.gen.ma.us>
			The Herd of Kittens