Part of the problem with polls is the defining of terms, and the Pew poll is an example of the problem one can encounter. To have "agnostics" and "don't know" separated doesn't make sense since "don't know" is being "agnostic". Even with "atheist" one has to watch for how this may be defined because I know of people who say any disbelief in a deity qualifies as "atheistic", which then would include "agnostic", "don't know", etc.
I have also seen polls that don't go along hardly at all with the Pew poll, quite possibly for another reason, namely how we may be referred to versus how we may refer to ourselves. For example, I'm Jewish, but if asked I would put down either "agnostic" or "don't know" or "Judaism", I would have some difficulty largely because I do affiliate with "Judaism", but yet I'm non-theistic.
I don't really have the time to look it up, but there are two polls cited that very much go against the Pew poll, one cited by research cosmologist Leonard Susskind and another from another source that I simply cannot remember, but both point out that most scientists tend to be atheists or agnostics-- mostly the latter. Also, Susskind states that most of the research cosmologists that he knows that are theistically inclined aren't very orthodox with their beliefs.
Of the various scientific groups, cosmologists are the least theistic followed by physicists. Frankly, I can't remember which scientific group was on the other end of the spectrum, but I do remember that slightly over half said they were theists of one type or another.
Also, is the Pew poll of American scientists or all scientists? The poll Susskind cited wasn't just U.S. scientists.