I guess it comes down to which methodologies you trust.
Ars technica prefers netmarketshare.com:
http://netmarketshare.com/faq.aspx#Vary
Why does the usage share reported by different sources vary?
It could be a number of reasons. One of the key differences is in methodology of data collection and reporting: Is the report based on unique users or pageviews? Is country level weighting used? Is there a quality control process in place to filter out anomalous data? In addition, having a global coverage versus data collected on a few individual websites also yields drastically different results.
