not so turbulent thursday, page-11

Pier1, for data testing, you would be well served by doing some out of sample testing. So, if you developed the strategy by eyeballing price action in the 2000-2010 window, then its possibly a curve-fit, unless you can validate it on say, 1990-2000 data. If there was no 'designed for dataset' in your strategy then maybe its OK.
In my experience in this stuff, 500 trades should be quite a good sample, in fact I tend to think about 100 trades would be enough, so you've got ample more than that. As for optimum dataset size, well, optimum would be as much data as possible.
As for your second point maybe someone else could give you an answer better than me, because I always use fixed position size (no compounding). I always look at the equity curve over time, a non compounding strategy, you want as close as you can get to a straight upsloping line. If you add in compounding the equity curve becomes skewed as position sizes increase (P/L size and drawdowns also increase), hence bigger compounding will make things look better if the last few trades are wins and worse if losses. To put it another way, compounding will skew your results towards the last trades in sequence.