Bill123,
I suggest we back up a little here.
First of all, how many shots did you take to determine that your unfired brass SD was 4? Or that your fired brass SD is 9?
We need to understand what SD is really telling us before we use it to investigate our reloading process. This bit me when I was load developing for a 7mag and if I new then what I know now I could have saved myself some frustration.
I'm going to skip to the end and save some suspense and just tell you that with a small sample size of 5 or 10 rounds you really don't know what your true SD is for whatever reloading technique you are using.
To make a long story short, if you fired a ten shot string to calculate the SD, then mathematically we would have 95% confidence that the REAL standard deviation of your load would be somewhere between two thirds and approximately two times your sample SD. So if you fired a ten shot string to collect your sd data, with a perfect chronograph, with 95% confidence the true SD of your unfired brass loads is somewhere between about 2.5 and 8. It could be anywhere in there. For the fired brass with same assumptions it is somewhere between 6 and 18. It is quite possible that the fired loads do have a true sd that is better than the unfired. If you only used five shot samples, then we have far less confidence since the distributions we would expect would overlap even more.
We haven't even addressed chronograph error, since that will reduce our confidence (and is certainly present). Also an outlier can really skew our sd calculation in a small sample size.
Bottom line, statistical tools like SD really don't tell us much until we get to at least a sample size of ten, and even then we can only estimate that it is within a range of about two thirds to approximately double the SD calculated from that sample.
This is the reason that 3 shot groups can lead you astray as well. We shouldn't read too much into what SD is telling us for small sample sizes.
I found experimentally that the real sd of my 7 mag load is better than what I first measured with a small 5 shot sample, and I wasted a lot of time trying to improve what wasn't really broken. We shouldn't have too much confidence in either a really good or really poor result from a small sample size.