The decision to go with PostgreSQL over Cassandra (or another distributed system like Riak or HBase) was simply because it gave us the most flexibility to change our product quickly while we operate at low scale. And if I'm honest with myself, we're operating at very low scale right now.
In the future as we scale up, Cassandra's distributed counters will be one of the first places we look.
Just my guess, but I'd be surprised if it's orders of magnitude different.
Scale up first. Use redis or memcached to store 'overview' stats. If you lose those stats, recalc from Pg.
While we're here, here are some PG tips:
1. Make sure your PG install is fully optimized. I've seen countless postgresql.conf's at major companies with no opts what so ever. Take a look at this book: http://www.2ndquadrant.com/books/postgresql-9-0-high-perform... .
2. Make sure you're using PG 9+.
3. Use pgbouncer for connection pooling and try to write your app so you can use 'pool_mode = transaction'.
4. Please don't run PG virtualized.
Have you looked at distributed counting in Cassandra for your counting needs yet[0][1]? Great info on its development and use at Twitter. It seems like you too have lots of interesting things to count. Your initial choice of Postgres for everything was a bit interesting. Your problem seems like a perfect fit for a hybrid solution (which you are already implementing by way of redis) that I think more and more companies will come to embrace by hook or by crook.
Continued success!
[0]General nosql at Twitter and a fair bit on Cassandra: http://www.infoq.com/presentations/NoSQL-at-Twitter-by-Ryan-...
[1]Specifics on Rainbird, the counting system at twitter built on Cassandra: http://www.slideshare.net/kevinweil/rainbird-realtime-analyt...
[cross post from the blog]