It might be easier if you broke your problem into several simpler parts, and then to combine the parts. For example, you could break down your problem and generalize it like this:

• Normalize the values of each attribute into a common set of units.
• Combine the values of all of one person's atributes to get a total for each group member.
• Sort the members of the group by this attribute total.
• Choose the top X members of the group.

Algorighmically, the only thing here that isn't O(N) is the sorting. That's not so bad.

On normalization: It really depends on the nature of your attributes, as to what kind of normalization makes sense. One easy way would be to arbitrarily assign the maximum value of any one attribute the value "1", and scale the rest of the values of that attribute from 0 to 1. But again, this probably wouldn't make sense in all cases- certainly not with heights and weights.

As for the perlish aspects: I'd recommend storing each person as a hash, with each attribute name and value as a key-value pair in that hash. That way you can add more attributes, or data about the attributes (normalized values, sums, etc) into each person without adding more and more arrays into your code. Other than that, once you decide on a solution post some code and I'm sure someone will be happy to help :)

Alan