Using this approach expresses a cynical view of troublemakers: They must be beyond any help, for we don't even want to discuss our banning them. Furthermore, you are assuming that they will not figure out that the delays are related to their username or IP. Finally, you are assuming that they don't know about your stealh-moderation technique, relying on SecurityByObscurity.
As for the non-ethical aspects of this: The problem will be outside channels of information as remarked by readers of Tom Coats' article. When the stealth-moderated users finds that only he is suffering "technical problems", the conflict may escalate. I don't know whether the affected user's reaction will be worse than after him being banned.
Also from the comments on the above article, your "good" users might either loose trust in the system when they hear about abysmal performance via outside channels, or if they are suffering real technical problems, they might feel you banned them.
This is a form of automated DissuadeInteraction.
You could also do this openly - if you have a rating system (eg karma) on your wiki, you could prioritise server requests by karma - if someone has a high karma, then you'll serve a request from them before a request by someone with low karma.
It seems to me that Tom Coats expects gaming wherever there are scores to gain. The reason stealth-moderation works, it would seem, is precisely the fact that the ones affected ignore the reasons for the delays they are experiencing. I think this illustrates the limits of stealth-moderations: It only works on ignorants. But then again, many moderation systems rely on simple barriers to entry that only exist for the casual spammer or careless newbie.