I agree - a good response to the original question is to recommend the biggest, fastest computer(s) available on the market today. Realistically, though, this is not feasable, so the question becomes a business one, as in "What hardware do we think the customer will buy?" From a development role, the important thing is to make sure that the software runs acceptable on the system that is recommended (with mod_perl that probably means a lot of RAM). Also it is useful to recommend a configuration that will be easy to support.

Given all of this, it is probably best to take this question from a business/sales standpoint and ask the question "what is the best way to make money off this recommendation?" Spending a lot of time benchmarking can eat into profits without providing a tangible value for the customer, so in general it may be best to test as few configurations as possible.

Also, a good solution might include suggesting lots of horizontal scalability - instead of a single server, why not recommend a horizonatally scalable solution using a load balancer and a couple of less powerful computers rather than one large server? This way, if additional hardware is required to meet performance expectations, additional (less expensive) servers can be added to the configuration. This configuration may also be easier to maintain, as individual servers can be swapped out, upgraded, etc without causing a system outage.