A web farm is only as strong as its weakest link. If any part of the architecture fails and is not prepared to handle that failure, the entire architecture fails. A fully resilient web farm needs to be planned all the way through.
Each item has a different failure rate and also a different cost to make it fully redundant; therefore, the cost/risk decision needs to be determined by you and the decision makers around you. You may find that it's worth the risk to leave some parts without redundancy for less critical web farms, or you may have double or triple contingencies at each part of the process.
When planning your web farm, be sure to ask the following question: Do you have redundancy or failover options at network feeds from upstream providers, power, all network equipment, firewalls, domain controllers, DNS servers, web servers, content servers, database servers, session state, documentation, processes, and staffing?
You may have a requirement for geographical redundancy so that if an entire city were to be taken out by a natural (or unnatural) disaster, your websites would remain online. This often requires a complete copy of your primary location with equipment that sits idle, waiting for the once-in-a-lifetime failure to occur.
The point in all this is to make sure to sit back, put your feet up, and think through every part of your architecture. Simply having a web farm with multiple web servers doesn't make a fail-proof system. Your network team, ...