I agree with Mike and would like to go a step further. We identified a few places in the Alfresco 3.2 one of our customers is using where non-optimal implementation of Alfresco persistence (not a rant, just a matter of fact for every complex system) is directly related to horrible performance observed in certain scenarios. Especially the aggressive pre-caching/pre-fetching of child content employed is one of the direct reasons why spaces perform as you reported. I'd like to challenge a standard Alfresco with an optimized (read "selectively patched") one, observe the differences and identify potential dynamic bottlenecks for change or introduction of self-adaptability / configurability.
I am planning to do something like this for months now, but unfortunately there are only so many remaining free hours a day to do it.