This question of using lowest ES/SD versus group size has come up quite a few times in recent months. The "answer" is not necessarily simple and there are likely a number of factors involved.
For starters, the occurrence of "positive compensation" is very well-documented. Positive compensation is an effect where slightly faster-moving bullets exit the bore ever so slightly earlier in the harmonic cycle than do slightly slower-moving bullets. As a result, the muzzle launch angle is slightly higher for the slower moving bullets (i.e. slightly later in the upswing of the harmonic cycle). In other words, the muzzle is pointing slightly higher when they leave so they have a slightly higher arc and arrive at the target at nearly the same vertical height as the faster-moving bullets, even though they have a tick less velocity. The slightly faster-moving bullets exit the bore ever so slightly earlier when the muzzle has a slightly lower launch angle. I have asked a few shooters that test loads at 1000 yd about the process of positive compensation and they have suggested that the velocity "window" over which this process occurs is probably in the neighborhood of about 20 fps wide.
In my mind, where the disconnect occurs is that many shooters here often state that really low ES/SD (i.e. well under 10 fps) isn't necessary, often with an undercurrent (my interpretation) that it might actually be bad. Further, they often state that their very best groups/loads often have an ES in the mid-teens. However, as I understand the process of positive compensation, as long as the velocities of all the shots in a group fall within that ~20 fps "window", it shouldn't make any difference. The groups should still benefit from the process of positive compensation and exhibit less vertical at long distance than than would be predicted by a ballistic program solely based on the individual shot velocities. In other words, any load with an ES below about 20 fps, as long as it falls within the positive compensation velocity window, should show the benefits of positive compensation in terms of vertical. For example, it shouldn't matter whether the ES was 15 fps or 4 fps, as long as the velocities were within the window. So very low ES shouldn't necessarily be bad, unless it is actually out of the window. Either there is something we [maybe it's just me] don't fully understand about positive compensation, or else the velocity numbers are misleading.
I personally favor the latter interpretation. It is common to measure velocities in 5-shot, or sometimes even 10-shot groups during the process of load development. Such sample sets are nowhere near large enough to gain statistically relevant ES/SD data. In fact, if you obtain enough 5-shot velocity data sets using the exact same load, you will usually begin to see how much variance there can be in the ES/SD values between different 5-shot groups with the exact same load. In reality, most shooters don't generate that amount of velocity data. Who wants to burn out their barrel by the time a single load development is completed? So we typically shoot small 5-shot groups during charge weight and seating depth testing, look at the numbers, and try infer some statistical relevance to the velocity data using samples sets that are far too small to provide statistically reliable information.
Based on my interpretation regarding statistical analysis of velocity data and the small sample sets involved, my feeling is simply that you want the ES/SD to be as low as you can reasonably achieve. Clearly ES values in the 40 fps, 50 fps, or even higher, are generally not acceptable, nor should they be. First, with such a high ES value, the velocities of individual shots will clearly fall outside the optimal velocity window in which positive compensation occurs (~20 fps), meaning you will have excessive vertical dispersion. Second, it is not difficult for most people, with a little experience and attention to detail during the brass preparation and reloading process, to routinely obtain ES values under the apparent 20 fps positive compensation window. The same cannot be said for routinely obtaining ES values well under 10 fps, especially with certain cartridge/powder/primer/bullet combinations; i.e. the smaller the case, the harder it often becomes to keep the ES/SD that low.
So the idea is use good reloading practices to keep your ES/SD reasonable, and you pay very close attention to what the targets tell you in order to determine where the optimal load parameters actually are. Pretty simple, really. Velocity measurements can have their place in terms of quickly finding the approximate region, and possibly expediting the load development process, but ultimately, the target has the final say. Velocity data can also be used as a "quality-control" mechanism, in that of your ES/SD values are unacceptably high, you know you have a little more work to do.
