Hi S128,
You will find many different opinions on this, so I might as well add mine. I find myself somewhere between the negative views of davidft and the positive views of brigidmac.
I agree with David that Ancestry oversells the value of the ethnicity testing, but I still think it tells us something interesting and sometimes useful. The areas are a little arbitrary and overlapping, but that is because populations (and DNA) vary gradually, and do not exist in tight well-defined blocs. The reference populations may be small or arbitrary in some cases, but that are still useful information. The percentages do keep varying, so we shouldn't set too much store on the exact numbers, but they still give us a general picture.
I think one of the difficulties is timeframe. If, as is often the case, our ancestors moved around a bit - whether a long time ago or more recently due to war and persecution - then we may have near ancestors in UK, older ancestors in Germany and ancient ancestors in an arc across Europe from Turkey to Scandinavia (as is my case, revealed through mtDNA testing). The autosomal test will reveal relatively recent ancestry, but that may still be varied if we go back more than a few generations.
The reference populations will include this sort of variation, hence the fuzziness of the areas and estimates, and the differences between the different companies. I have ethnicity estimates from Ancestry, FTDNA and Gedmatch (which has many different estimates) and they are all different, but there is a reasonable commonality too. I haven't yet found ethnicity useful in developing my family tree, but I don't think it will always be irrelevant, especially as more and more people test and reference populations get larger and better defined.