Undoubtedly the bigger the sample size the more confident you are of your results. My system which just covers WA is based on results from 1997 to present day and the fluctuation on a year to year basis is 97.2% returned for 1998 to 112.8% returned in 2002 (Based on Tabcorp prices).
This means if for example I had tried the system out at the end of 1998 I would have binned it as useless because it had a slight negative return - and that would have been based on over 3000 bets.
As my system chews through the data it updates the percent returned on screen after each race and the (rapidly moving figures) for each year shoot up and down wildly for the first part of a year, they can even move about quite a bit as the year being processed nears the finish. This just illustrates to me the clustering of wins and losses and convinces me that anything less than a couple of thousand results (unless you are showing a very significant win or loss percentage) is pretty meaningless. My system is based on form and selects a lot of favorites so I'm not talking about a system that is relying on a few long priced winners.
When I started using real money about a year ago, the first week I lost 7 out of 7 races but because I had over 7 years results behind me I never doubted it would come back to profit. The more data the more confidence you can bet with.

With regards SP versus other divvies. My system returns negative 0.8% on SP over the last 8 and a half years and positive 5.26% on Tabcorp. I don't know if that's typical or if my selections are just closer to bookies selections than Tab ones. On maxidiv which is where I bet at present I'm laughing. Nothing like hearing a horse pay $5.00 when you listen to the race on the radio and seeing you got 7 or 8 next time you fire up the computer.

KV