dbSNP is very trustworthy in terms of clinical associations because the clinical associations seen on a dbSNP entry are linked to OMIM entries. OMIM is manually curated and the reports are generally rather exhaustive reviews of the literature.
I have too little experience with SNPedia to make specific comments. As a wiki, SNPedia has the usual advantages and drawbacks of any publicly curated site. The info could be expert (and timely) or could be junk or anywhere in between.
SNPedia would prefer if you didn't blindly trust it. It tries to link all information back to the primary sources, and encourages you to view those sources for their full context. However due to the "many eyeballs" which use SNPedia / Promethease and Linus's Law it is often good at summarizing large volumes of primary literature. It sources information from dbSNP, OMIM, HapMap, PubMed, genome.gov, PharmGKB, self reported Direct to Consumer phenotypes and many other sources. It tries to complement, not compete with all of those sources. New papers are added every day, and when conflicts between sources are discovered by anyone they can be documented publicly and immediately, instead of waiting for a more formalized reporting process. All changes are visible and attributable in the wiki history. A paper better explaining SNPedia has been accepted for the next database issue of Nucleic Acids Research, but email@example.com (or direct edits to SNPedia) can be used to raise further questions.
A recent example of the benefits of this system is from a Promethease user who noticed a conflict between PharmGKB and most other sources about which genotype is sensitive to caffeine. See <http://www.snpedia.com/index.php/Rs5751876> for the specifics.