The solution proposed by Fletch is what I would do, outside of moving to the object-oriented approach.

If you're talking absolute speed, then maybe you'd want to stay away from an object-oriented solution, but I think in practical terms, especially considering the huge DB overhead, the measurable difference in terms of performance is going to be 1% or less, which is to say, external variables will have more of an effect (i.e. CPU load, network traffic, etc.).

What I mean to suggest is that just because an object oriented technique is "slower" doesn't mean that it must be avoided. "Fastest possible" is certainly not meant to be taken at its most extreme interpretation, because if that were the case, you can forget about using Perl at all.