I've found that this problem happens less, as Jeff observed, for common queries (e.g. "Create GUID") but happens a lot for less-common or complex queries, where relatively few questions contain all the keywords but several related questions' titles, when combined, do match all my keywords.
So instead of getting one or two highly-relevant SO results from Google (or zero results, letting me know I need to change my Google query), I'll get 50 results of which only a handful may actually contain my desired keywords, and the majority simply contain links to lots of related questions.
Also, in the "abstract" text shown under the search result, Google often chooses to include titles of related questions instead of actual question or answer content. (This makes sense since the related titles have my desired keywords while the question or answer does not.) So not only do I get irrelevant results, I also can't read enough of the question or answers of that result to know it's irrelevant. I need to click through each one to find out that it's not relevant.
The net impact: I need to spend much more time hunting through Google pages and irrelevant SO pages before I either find my answer or change my Google query to get more relevant results.
An example of this is searching SO for flash video redirect streaming rtmp site:stackoverflow.com. A screenshot is below. I added red boxes showing titles of related questions, and the green boxes are actual text of actual questions or answers. As you can see, related question titles dominate the result summaries. 4 of the top 6 results have so many related question titles clogging up their summary text that no text from that question/answer is shown at all!
That said, I'm not fully convinced that Google indexing related questions is 100% bad-- because this "feature" might cause me to find results on Google results that I might not otherwise find, in cases where I'm searching for a synonym or similar term which is never mentioned on the page itself.
It might be worthwhile to run a test like this:
- pick (e.g. from web logs' referrer URLs) a set of difficult or representative search queries for questions matching a particular tag. Record how many results Google returns and how relevant those results are. (where "relevant" means do the top 3-5 results contain, subjectively, the "right answer")
- Run the test for a month or two to allow Google's indexes to catch up.
- Re-run the same queries and see, subjectively, whether the results (and each result's summary text) make it easier or harder to find the right answer in less time.
- Decide if the decrease in noise outweighs the (perhaps theoretical?) decrease in relevance for synonyms/related terms.
Jeff, I suspect if you'd be willing to do step #2 above, I bet that the SO community could help with the testing part.