PostgreSQL pseudo cluster

I am trying to build a semi-scaleable (up to three SMP servers),
highly-available database without forking over $300k to Oracle. I expect the
app to be about 90% reads and 10%writes. The architecture will be the
Internet into a Cisco Local Director load balancer that will forward
requests to one of three web servers running Apache/PHP. The web servers
will have redundant Gb Ethernet cards running into redundant switches, which
then run into the three database servers. Each web server will have it's own
dedicated database server for reads. For writes, I was planning on having
wrapper functions that would modify all three databases for each modifying
transaction, thus keeping my data synchronized between database servers.
Each database will have it's own storage, as most of the data for the
application will be graphics files that I will keep on a NAS, reference by
the database instead of storing in the database. Does anyone have experience
with such an architecture? Does it sound prudent?

Thanks,
Mike Millsap