Firstly, they have a clear conflict of interest, which has been discussed many times, when it comes to selling advertising. Buying advertising (anecdotally) seems to make bad reviews magically disappear.
Secondly, and this has always been the problem with "local", is you need a certain critical mass for it to be usable. You can argue that Yelp has reached this point in many cases (although see the next two points) but there are many businesses with <5 reviews.
Third, there is too much friction in asking people to review (and even rate things). Most people simply don't and probably never will. This exacerbates the "critical mass" problem but also introduces a selection bias. The people who comment and rate aren't necessarily representative of general opinions or you (the personalization problem).
I've gone to eat at some places in NYC that are 3.5+ stars that have varied from average to terrible. In some cases I've gone with someone who shared this positive review but--and I realize the counterargument to this is that it's subjective--they're just wrong.
Fourth, there is a clear fraud problem with reviews and ratings. People are clearly paid to give positive reviews (eg you see someone rate a given car dealership in the Bay area on one day and then another in Maine the next day and so on). Of any of the companies in "local", IMHO Google is in the best position to deal with this particular problem (disclaimer: I work for Google).
Lastly, as such reviews become increasingly important, there is the issue of extortion. If this hasn't happened already it will. Criminals already target websites with DDoS attacks that go away if the site in question pays what amounts to "protection money". There's nothing really to stop such criminal enterprises shaking down businesses with the threat of a bad slew of reviews.
It's worth making extra mention of personalization. Many (Google included) seem to view "social search" and "social recommendations" as some kind of panacea to some or even all of these problems. I disagree. I know a couple of people who, say, like Adam Sandler movies. I do not. Not even remotely. Their movie recommendations are so diametrically opposed to mine that I can pretty much take the opposite of what they recommend.
The way forward with this will be something like the Netflix model (IMHO) where these great data mining systems will attempt to find people who are like me and have similar tastes whose recommendations will likely coincide with mine.