It feels like a cheap cop-out for controlling spam: the side effect is that you have heavy hitters like Maholo, Target, etc... all out spamming the more relevant websites.
A more general example of this algorithm problem is that it ranks the pages of Digg, HN, or similar to be higher than the actual website it links to, and most of the times those pages are devoid of any actual content, merely containing a couple of votes and no comments.
A guess is that this is the result of the reliance on domain names for ranking. The high-quality lyrics are distributed across millions of separate domain names, since fansites typically are dedicated to only one band. The low-quality, ring-tones-heavy lyrics sites, though, are centralized, with every band's lyrics on the same domain.
Expected result is something like: http://www.joydiv.org/shadowplay/joyd/warsaw.html
What I actually get are a bunch of sites with names like mp3lyrics, songlyrics, musicsonglyrics. It might be that the word "lyrics" in my query is boosting that sort of stuff to the top? From what I can tell, Google also penalizes old / rarely updated sites, which probably hurts these kinds of pages--- the page linked above was last updated in 2007, but it's still the best result for that query.
I was thinking I could email you at the address in your profile, but when I think about it, it might be more productive to post a Ask HN here once I have a better explanation of what I mean. Then the whole community could participate, which seems like it would be constructive.
Also, to be clear, I'm speaking in an unofficial capacity here, as myself. I'm quite curious what people's frustrations are, but I can't promise anything.
Google has let their core brand go bad.
When you do find a lot of spam, does Show Options -> Fewer Shopping Sites help?
I work in Google Search, and we actively try to identify areas that people are having difficulty with and then come up with ways to make their search experience better. Unfortunately, most of the low-hanging fruit has already been picked. If we knew precisely where the pain points were, we'd be much better able to solve them.