I read Jonathan's article. Disappointing that such a basic counter is suspect. And even though 470GB of memory sounds like a lot, it's only 6% or so of the 6+ TB of data. Management seems convinced that we're safe to proceed with the SSD rollout to production, but this mainly OLTP testing that was done was off the mark. Report queries with lots of IO are our problem.
I suspect if you took out lots of index/statistic maintenance the writes would be lower
Test environment logical reads during SSD testing window
Test environment physical reads during SSD testing window
Production logical reads during weekday morning hour
Production Physical reads during weekday morning hour