midi is light but VST's are very heavy. Be wary of multiple cores. Many VST's can only take advantage of a single core. So if you have a VST that is particularly heavy, it might not matter how many cores you have, but the performance of a single one. (in that case a single 3GHz thread would be better than 3 2GHz threads :S)
But whatever way you go, often what is as important, if not more important than CPU is the latency of your soundcard/interface and how much RAM (and its speed) you have. Low latency soundcards can accomodate higher buffering with relatively small delays. The more RAM you have the more you can 'preload'...i.e. if you CPU is failing (your song starts to go jittery etc), increase the buffering (or number of samples or whatever it is called in your soundcard) so that more of the song is loaded into RAM to start with...the higher the buffering the larger the delay in actually hearing the music, but if you are mixing/just using VST's (i.e. not recording) it doesnt matter.
Typically, when recording I'll set the buffering to as low as possible so I can get things in time (because if there is a delay in what I'm hearing, everything I record will be consequently delayed :S), then, when mixing and using CPU heavy plugins etc, I'll shove up the buffering as far as necessary (typically 512 or 1024 samples) which introduces a delay of around 10ms. Doesnt matter when mixing though as everything is delayed the same amount
And do what James says if you buy - go for a 'stripped down' computer.