As an Ag Economist, I'd say I am fairly well based in statistics, but that is beside the point.  I would assume you are familiar enough with 'demonstration plot' practice to know that there is a world of difference between planting whole fields in alternating strips with two different hybrids and how the vast majority of so-called plots are done.  There is much room to defend the former, although it can be biased if the operator has an agenda or simply doesn't know or think about what he is doing.  There is very little room to defend the latter. However neither can produce the data which can come from replicated, randomized trials.  This remains the "best design" at this time, and I'm willing to bet will continue to remain so for the forseeable future. 

Just when did the goal become to "average the spacial variation that exists in any field?"  The goal, as I understand it, is to judge the variation between the hybrids, eliminating as much outside variation as possible so that you can get a std dev low enough to actually predict something.  We want to evaluate hybrids, not throw a million other variables into the mix to dilute the data with a bunch of non-controlled limiting factors which are neither identified nor measured, and which throw the confidence interval so wide that all you can really say for sure is that it was a corn field. 

If you are producing plots which require different harvest dates to accomodate massively different harvest M.C. it would appear to me that you are trying to compare hybrids with different maturities.  This, in my conditions, is going to do nothing but prove what maturity (more accurately, what polination date) was best that year.  Information which is approximately worthless because one year is very unreliable as a predictor of next year's weather.  We have to compare like hybrids and depend on other, very long term, data to pick pollination date. 

I suppose, after reflection, that geography has something to do with outlook on this topic.  In my geography, about all you can ever learn from statistical analysis of common demonstration plots is that they have a tremendously large std dev.  Maybe corn belt guys can get away with sloppy plot design and still learn something.  In my experience, however,  you can glean far more useful information from the yield maps of a so-called plot (which is done as most demonstration plots are) then what you can ever get from the weigh wagon.  In order to do this, you have to use a fair bit of judgement to pick limits to throw out data.  The inevitible result is that the product is very hard to defend as it is greatly based on the skill and judgement of the person producing it....hardly a scientific process yet it still can produce useful information if you are doing it yourself or have a trusted source to produce the judgement calls. 

Sorry this ended up so long.  It's late and I'm not going to redo it for conciseness.