fukuyama

Personally, I would choose UTA. There are two really good faculty there in my area (data mining). There actually isn't a good reason I didn't apply there to work with the IDEAL lab. I have had some contact with the professor before as a Master's student. JHU has never been on the map for me, but it might depend on the area. I would take a look at IDEAL and DML labs there. Really good faculty (personally, as well).
 
I think it's such a personal choice, and there are a lot more important factors to consider than ranking. I dont think the exact ranking matters much, but it is important to be in at least the second tier schools (i.e. top-100 in US). In top conferences you'll be interacting with people mostly from these places. As for myself, there isn't anyone I follow at an MIT or Berkeley, so those rankings are meaningless.

You have five good choices. I think the choice of school isn't as relevant as factors such as faculty pairing, and lab environment. 
 
My colleague has a really good post here: http://www.theclimatecode.com/2013/02/a-phd-is-like-haute-couture-its.html . I think the best (if you can manage) is to visit the labs you are interested in with these things in mind, or even email some of the senior PhD students to get a sense. All of them are within the top quarter of universities in terms of rankings so your opportunities won't much depend on 'name' recognition. In any case, places such as MSR or Google Research are less focused on pedigree than in tenured academic positions. You'll find several people in Machine learning and data mining coming into these places with Hong Kong, Greece, Singapore PhDs, or from lower-ranking universities in the U.S. and Canada. 
 
Another note of advice is that in choosing an advisor you should keep in mind that your advisor does not dedicate themselves to your goals. At best the advisor relationship is mutually productive, but (especially with 'superstars') it can be somewhat predatory; having a strong idea what you want to do and what resources you need to do it will save you months or even years of 'working for' your advisor, and being pushed out with a lackluster publication record because of expected PhD timelines.
 
I'd contact three or so faculty at each of these schools, looking for a match. Also consider that redundancy is a good factor. My M.S. was at Minnesota, where there are a good selection of data mining faculty (5, solidly). People switch advisors if things dont work out, and being at a university which is an island of you and your advisor is a precarious situation.  

This isn't a very precise question. I would say most large CS departments have an intersection with classified research. There are government labs (Oak Ridge National Labs), government-funded university labs (MIT Lincoln Labs) and there are projects which either have clearance or citizenship requirements (or both) which either restrict from on-site visits (i.e. even some non-citizen PIs can't see the end implementation on-site) or faculty and students who work on the project (DARPA has a lot of these). 
 
My undergraduate and master's university (Minnesota) had relationships with all of these. In my area (data mining), clearance might open a few doors at ORNL, but I dont think it's a driving motivation or advantage. As a U.S. citizen grad student, I have been approached by at least three faculty who want to write a DARPA proposal with me because I meet this requirement. With clearance, the solicitations might be more. 

This isn't a very precise question. I would say most large CS departments have an intersection with classified research. There are government labs (Oak Ridge National Labs), government-funded university labs (MIT Lincoln Labs) and there are projects which either have clearance or citizenship requirements (or both) which either restrict from on-site visits (i.e. even some non-citizen PIs can't see the end implementation on-site) or faculty and students who work on the project (DARPA has a lot of these). 
 
My undergraduate and master's university (Minnesota) had relationships with all of these. In my area (data mining), clearance might open a few doors at ORNL, but I dont think it's a driving motivation or advantage. As a U.S. citizen grad student, I have been approached by at least three faculty who want to write a DARPA proposal with me because I meet this requirement. With clearance, the solicitations might be more. 

This isn't a very precise question. I would say most large CS departments have an intersection with classified research. There are government labs (Oak Ridge National Labs), government-funded university labs (MIT Lincoln Labs) and there are projects which either have clearance or citizenship requirements (or both) which either restrict from on-site visits (i.e. even some non-citizen PIs can't see the end implementation on-site) or faculty and students who work on the project (DARPA has a lot of these). 
 
My undergraduate and master's university (Minnesota) had relationships with all of these. In my area (data mining), clearance might open a few doors at ORNL, but I dont think it's a driving motivation or advantage. As a U.S. citizen grad student, I have been approached by at least three faculty who want to write a DARPA proposal with me because I meet this requirement. With clearance, the solicitations might be more. 

As I understand it, in CS rankings are not reliable at all, for a start because most use journal articles (and surveys) to measure impact, while CS is very much conference-driven. Depending on what you're looking to do (i.e. an academic position) there are 'top' universities that help here as long as your record is also productive. But many people from middle-tier universities still get professorship positions. 
 
H-index by department isn't a terrible measure of activity within a research area. Microsoft Academic (http://academic.research.microsoft.com/) tracks this, and for my area of Data Mining it's a very plausible ranking (http://academic.research.microsoft.com/RankList?entitytype=7&topdomainid=2&subdomainid=7&last=10&orderby=1). For my area, the traditional top-ranked schools aren't much use (MIT, Berkeley, etc). 
 
I think far more important is finding the right advisor. I've seen people struggle for years with a poor fit of an advisor and it's a disaster: you dont get to work on what you want, you aren't developing your skills as well as you should, you probably aren't enjoying your work, and you probably aren't publishing at the level you want. I'd much rather be at a lower ranked school with a supportive advisor.