My startup has similar data volumes and reporting needs to rescuetime, and also runs rails on the backend. We found that traditional data warehousing techniques, and specifically running a MOLAP parallel to the SQL database, solved the performance problems so soundly that we could put 10x as many charts on a page and enabled data-slicing possibilities that were previously out of the question with the highly tuned SQL route. We were also able to get rid of all of the extra model code we had written to manage aggregations. See http://en.wikipedia.org/wiki/Online_analytical_processing for more background info if you're interested.