I recently switched to Giga 3 (Orch version) but I am having trouble getting a good polyphony setting. I have the max number of voices that can be used simultaneously at 400, but, for some reason when I am scoring large pieces (full orch or even just large string section) I am losing voices like crazy. For example, I was scoring a piece with violins 1&2, violas and basses all pizz and cellos legato doubled by bassoon and the long line of the cellos and bassoon would cut out each time after a few seconds. The number of voices used for the remaining strings pizz doesn't seem like that many - what give2?
If your machine is not capable of sustaining the 400 voices, setting the polyphony that high actually harms the performance. How? Because wherever you set the polyphony "counter" is where the polyphony-management system begins to look for voices it can free up transparently.
So, you may get a more smooth "apparent" polyphony at 200 voices than at 400 voices.
Other potential issues: Perhaps you're pulling some voices from a fragmented part of the disk, or a slower drive?
With GS 2.5, your polyphony management system was kicking in at a maximum of 160 voices. 400 voices is a LOT more load for the computer to carry, therefore the "cut-outs." So you might want to start at 160 voices, and creep upwards until the system begins to cut out. Once you hit that point, back off a notch, and you should be running smoothly.
I had some polyphony problems with GS3.0 that was related to the fact that I had rewire enabled (while used as stand alone), just make sure rewire is disabled (in options / system), this might help getting a better performance.