Do you run CQ long enough to force the weaker ones to leave?
The term 'clear' frequency is relative. The important point is not to cause interference to anyone else. Quite often you can coexist on overlapping TX bandwidth with other stations and not cause interference to them. A good practice is to ask if the frequency is in use a couple times, then if nothing heard, blast away. If you get asked to move in the first few minutes, then move. But if you get a good run going and no one is complaining, keep it up. Quite often if you have occupied a frequency for a long time, propagation will change and you will get new complaints from other stations who claim to have been on the frequency for hours; this can easily be true but as propagation shifts, interference issues become apparent.
Should you limit your TX bandwidth to take as little space as possible, or should you widen it (with some overdrive), to discourage cohabitation?
Use a reasonable, normal TX bandwidth, like 3.1K. Do NOT occupy more than is needed in an attempt to drive others away. On a busy band, often narrowing the Rx bandwidth down to 2.1 or 1.8kHz is needed to cut out adjacent stations, either due to overlap or splatter. Imagine how much more space there would be if everyone reduced their Tx bandwidth to 2.1K! With old crystal filter rigs this was impossible, but with today's DSP rigs it is more than doable.
Or is it best to occupy a frequency at the start of the contest and never relinquish it until the propagation is over?
Unless you are one of the biggest stations on the band, this simply isn't possible or practical.
My station in Georgia runs a KW to 2 phased beams at 65' and 100'. It seemed that stations on the East Coast with modest antennas were doing much better.
Not surprising; there is a definite advantage to being on the East Coast with one less hop to Europe. But your power and antennas should have done fine. Note that conditions were pretty poor, especially Sunday, and even here in 1-land the loudest EU stations were not more than 20 over, would have expected some 40 over.
One culprit could be Flex 6600 where setting of mic gain with speech processor is critical. With Flex I experience many more repetitions than with K3. But I had good runs on 80m (4 square).
I highly doubt it's the speech compressor setting. The Flex uses CESSB which is excellent at increasing average power and also very immune to being overdriven. It's much more likely it could be a Tx equalization issue. You want to use Tx equalization to 'fill up' and flatten the available Tx bandwidth. If you voice has a lot of low-end peaks, you might need to add 10 db of boos at the top end.
-Tony, K1KP