Well, lets get some facts. Feel free to downvote me.
First, you say that "the good questions that pop-up amidst the morass of rubbish get buried very quickly". In the first 3 pages of the newest questions, I see only 2 code-trolling questions, one is near the bottom of first page and one in the third page. Don't think that 2 questions far from the top buries the "good questions" very quickly. As usual, most questions are code-golf. From those code-golf questions, some are good code-golfs and some aren't, i.e. exactly in the way it always has been before. So your statement is not true. You might argue that this happens because we are getting more people posting here and good questions are buried, but this too means that new good questions are popping and that this site is showing some signals that we are finally getting out of beta, and I think that this is good. A proof of that is this question: Produce the number 2014 without any numbers in your source code
We recently had three questions that were explosively upvoted and answered:
Three of the most voted questions in only a month, including the first and the second place. Surely my question was the most explosive ever in the site, but the third one (a very good code-golf) is the second one, currently is the hottest question in StackExchange and was posted only 4 days ago. Do you not think that this question would not attract a lot of inexperienced users too?
And, you are looking only to the questions, not to the answers. Truely, some answers are poor, exactly how it always has been before the day that I invented code-trolling, but some answers are very good ones. Personally, I learnt a lot in this site coding something that does something interesting and looking for creative and original answers that features creative and original ways to do something than looking byte-crushed illegible golfscript answers that serves for no purpose and that teaches me nothing. And this is the main reason why I think that code-trolling is valuable, I could learn for fun and laugh, and a lot of people too, and I think that learning while getting fun is the real purpose of this site. Even some very poorly conceived code-trolling questions got some really creative and original answers.
But now, we have a lot new users, thanks mostly to those three questions. Surely, there are some bad new users, but there are good new users too, and if you gets new users in any StackExchange site some are good and some aren't no matter what. Bad users will not get much upvotes, and will either go away quickly or will learn the way to become good users, and most good users stay. Sometime before I posted my question we had around 4 questions per day in area 51 and were declining. Now we reached 7, which I think is excellent. You could argue that new users should come more slowly (and if you does, I will strongly disagree), but this is no reason to delete code-trolling questions.
In fact, you are using code-trolling as a scapegoat and not addressing the real "problem": We simply have more users, a lot more users. Deleting code-trolling questions will not solve your "problem". In fact I don't see this as a problem after all.
And about the the "dramatically fall in the quality of the questions", again, only 2 questions in the first 3 pages are code-trolling, so code-trolling is not the problem here. In fact, I think that this is a misperception caused by the wave of new users. What I see is that some new users post some poorly conceived questions frequently missing a clear winning criteria, and code-trolling has nothing to do with this. Deleting code-trolling questions will not change this either. Surely, some people post low quality questions (as they always did, whatever the tag is), but this is the reason that we have the downvote button.
We have only 16 code-trolling non-deleted questions now, only 8 with more than 1000 views. All of them pretty quiet now. Don't think that deleting them will magically improve the quality of whatever is the overall quality of this site.