There are different angles on this, and they vary depending on the field and subfield. But I think it would be fair to say that t
here always has been demand for MS level statisticians, and there will always be. From an employer's angle, what matters is what you can bring to the organization. An opinion that I heard a couple of times that is really hard to argue with is, "There is little in Ph.D.'s tollbox that isn't in the MS toolbox". Unless you go to the company that has a fleet of 20-30 Ph.D.s, and is known to work on the projects that eventually end up in a series of academic publications, including a couple in JASA/Biometrika, you are being hired to do rather specific tasks (these series of cross-tabs, this type of power analysis for a proposal, this time series of customer satisfaction or student evaluations, etc.), and very few of them would require the development of new methodology (which is what Ph.D. students spend most of their time polishing after the first couple of years, and grow to think that statistics is all about this new methodology and publishing JASA papers).
Second, there are formal job requirements, and you have to look at the ones in the field that you are interested in to see the breakdown of MS vs. Ph.D. There are many positions (certainly in academia, but also in pharma) where Ph.D. is a requirement, so you simply can't apply if you don't have that piece of paper. In the line of work I am in, I have taught myself 90% of what I know in survey statistics, so my Ph.D. is fairly nominal, and measure theoretic probability does not cross paths with me, no matter how much I loved the subject, the course and the instructor I had in grad school. (Survey methodology was something I was learning outside of the statistics program, and that part of education turned out to be more relevant, even though an order of magnitude less technical than the graduate courses in theoretical statistics.) I wouldn't be surprised to find people with MS degrees and titles similar to mine, as there simply aren't enough programs producing Ph.D.s in survey methodology (unlike the dozens of biostat programs stamping a few hundred Ph.D.s a year, and creating an intellectual overhang on that submarket). On the other hand, whether Ph.D.s are overqualified for MS positions is a judgement call for every employer (see the point above about the job requirements and the skills that will or will not be usable; there have been different opinions voiced already in this exchange), and hiring and keeping Ph.D.s on MS level positions is only sustainable in the long run if the overproduction of Ph.D.s continues into the future.
Third, there are certain levels of task complexity, especially on the interfaces with other fields, where just the MS is likely to be unrealistically little (unless it is from a top university). If you want to do Bayesian computing for a job, or some cool *omics stuff, or "big data" with a very heavy computational component that requires a professional grade Java/Python/Ruby programming, then you would probably have to get a Ph.D. just to have enough time in school to get immersed in these multiple disciplines (assuming say one CS class per language, which is likely to be insufficient, anyway, and/or a bunch of biology classes). You would also need to have more of the Ph.D.-type frame of mind to learn to learn the new technological stuff that will be appearing every three to five years. A Carnegie Mellon MS in computational statistics, on top of CalTech BS in computer science, will probably do fine in a Silicon Valley startup, but a less stellar constellation of degrees will raise doubts of whether you simply has had a chance to learn enough to work in these multidisciplinary fields.
Fourth, there's general deflation of degree credentials. Some thirty years ago (when I was in grade school... not in grad school), it probably would've been fine to hire an MS statistician to lead a university consulting center, as there weren't that many statisticians around, and MS was a solid degree. When I looked at Mardia, Kent and Bibby's multivariate book that I used to teach a doctoral multivariate class from (in combination with Anderson and Hastie-Tibshirani-Friedman though), the book, written in the 1970s, placed itself as an upper undergraduate. It might have been then, but it is certainly a graduate level text now; and most doctoral students in social sciences would be greatly challenged by it. Due to the combination of gradually declining education standards and larger fraction of the young adults getting their degrees, college and above, you may have to have a higher level degree just to stand above the crowd -- to send a signal about your intellectual ability, as a labor economist would say (and may continue to say that this is a very costly signal: to you personally in terms of the number of years you have to spend in school just to prove that you are smart, to the companies hiring on the market who have to sift through the larger number of applicants, and eventually to the society as a whole). If everybody has a BS now, you have to get at least an MS to show off that you have better skills than an "average" fresh-from-school statistician; if everybody has an MS, you have to get a Ph.D. -- but then we are back to the square one re: the actual skills, as discussed in the first point.
Finally, some folks go get a nominal Stat MS degree just to learn enough SAS to get themselves a SAS programmer/data management/report prep job, with little if any statistical content. Again, there are jobs that require just that, and since calling SAS a programming language, let alone a modern programming language, is a bit of a stretch, this won't be something you would learn in a CS program. Instead, you would take a (bio)stat class titled "Statistical computing" or something like that, and would have to build up your SAS skills through both doing practical work in a statistical consulting center as a graduate RA, and collecting the various certificates through SAS training courses. Similar considerations apply to Stata and R, although building up R-no-statistics resume probably won't get you a job as easily as building up SAS-no-statistics resume would, and the certification/training is not nearly as structured. Stata users can just read J Scott Long's "Workflow" book that contains enough wisdom to count as a solid foundation for further practical work :).
These are my personal opinions based on what I've seen on the market, and what I've heard from other people who have been on both the hiring and the being hired sides of the market.
-------------------------------------------
Stanislav Kolenikov
Senior Survey Statistician
Abt SRBI
-------------------------------------------