Since this is somewhat of an opinion-based topic, I'll speak from my
own experience. Downtime where I am at is greatly frowned upon -
because of this I also favor using redundant, smaller servers.
In addition to reducing downtime, this will help you if you need added
capacity - you've already forced yourself to think about scaling and
as a result can just tack on added machines as needed.
The two most difficult areas to scale (in my opinion) are the database
and file storage. If you can solve these two issues then I see no
reason why smaller servers wouldn't suffice and in fact be beneficial.
I have had success running a centralized NFS store over a backend
gigabit network on a RAID5 server w/ high speed SCSI disks. The
database replication I have been toying with has been MySQL's