Wouldn't having "max_connections" postgresql processes create more problems?
What we see in practice, though, is that people often increase max_connections to rather insane values. There's only a certain number of active connections each box can support (say, ~2*cores, sometimes more), and it's one of the things we check whenever a customer contacts us with performance issues. Oh, you increased max_connections to 7243, it was running fine for a while and then you got a short burst of activity on Monday morning and it's crawling out of the rack? pgbouncer FTW in most cases (in transaction pooling mode).
We hit the max connections limit in postgres too at peak times (but no lock-up or similar thing happened) and per advice of our hosting provider we added pgbouncer to the stack last week and will hopefully don't have much problems here anymore.
Disclaimer: I wrote that article.