I agree the data side
In other words, instead of calling interfaces provided by these
backends (which is, obviously, more costly) , MyFedora will be
directly access the databases and manipulate the data to be in a more
user-centric and useful views (am I right?).
I generally do not want to talk to a DB directly though if performance
becomes an issue I might have to. Generally it goes
MyFedora <-json-> mfquery proxy <-json/xmlrpc etc.-> Fedora Resource
<-db connection-> DB.
The proxy is there so I can call async via JavaScript so the only thing
one would gain from a direct DB connection is to get rid of the HTTP
connection to the fedora resource but then you lose any of the business
logic above and beyond the actual data.