If hardware is so reliable these days then a good strategy may be to have clusters with one new (supported, warrantied etc) server paired with an old (unsupported, out of warranty) server. If you keep the cluster running on the old box until its hardware fails, cutover to the new server, replace the old out of warranty server with another new box then you may be able to let the capacity demands of the business on the hardware dictate the replacement timetable rather than a 3 year warranty cycle. You might find yourself replacing hardware when it breaks, every 5-8 years, rather than when the warranty runs out. After all, why would you fix something that wasn't broken?
(Edit: I have noticed that this article was originally posted a while ago!)