For the volume of mail affected by this on our network, CPU time is not worth mentioning. YMMV.
Very true.. It's going to heavily depend on amount of mail, and SA configuration.
It really depends on what the OP wants to accomplish, but I think "by far a better way" is debatable in the general case. :-)
Fair enough. I suppose "least prone to have potential for other problems" would be better...
There are other things than just bayes learning and tagging to consider in many sites... i.e: the AWL database (only a concern if all your mail is processed as one user.. but this is probably more often the case than not). The AWL is going to be a more common concern soon, since it's default-on in SA 3.0.
(note: the all_spam_to part won't poison the AWL, it processes pre-whitelist scores, but if you're wanting to avoid ANY kind of learning behavior for mail going to a given address, you'll need to consider it.)
But you are right.. "by far a better way" is subject to caveats of what the OP needs.
