You would use Octave/ Matlab instead of Julia, R or Python because Matlab syntax is the cleanest and most succinct way currently to articulate complex (in both senses) matrix algorithms. It isn't nearly as good for general programming, but it is much better than anything else for its niche. While some of that is obviously just fuzzy opinion, I am pretty sure you need fewer characters or LOC to say the same thing in Matlab than in the other scientific languages mentioned.
Matlab (and thus Octave) also has a HUGE installed base of engineers who don't really care that HN thinks the language du jour is X, Y, or Z and continue to use Matlab to get their work done (and will for the next 20 years). If you are working with them, you use Matlab. If you are working with them and don't have $1000 for a license, you use Octave.
Octave also has some syntactical improvements like "+=" that are missing in Matlab.
Julia syntax is mostly Matlab syntax with the parens-for-array-indexing replaced with square brackets. Its main shortcoming WRT Matlab is that it's a new language and a lot of things still need to be implemented, but the built-in linalg support is pretty good.
Python is also not that bad with pylab, which eliminates most of the verbosity of calling numpy/scipy functions directly, although it might still end up being more LOC than Matlab in some cases. I have only minimal experience with R, so I can't comment there.
So, I am back to Old Faithful. Really, I hate all languages, I just find Octave the least annoying in the mathematical programming space. (I used to love all languages, now I am old and bitter ;) )
Don't get me wrong -- lots of great ideas in all of the above languages. If Julia were to make vector / matrix thinking natural, I would switch.
Why not Python, etc? For the same reason that most matlab users would probably give: the research scientist that gives me code uses matlab.
If you had access to Matlab, would you use it instead?
Why not Python/R/Julia?
I know the Matlab syntax/commands/language and that enables me to be productive with Octave quickly. I just don't have the time to learn and evaluate Python/R/Julia (although Julia looks intriguing, but again, I just don't have the time).
edit: TL/DR - what forkandwait wrote.
As for Python/R/Julia:
Python is more general purpose than Matlab/Octave, though I have seen a push for NumPy and SciPy.
R and Matlab work in somewhat different fields (exceptions to the rule: Functional data people in statistics seem to like Matlab). I always got the feeling the math/linear algebra/engineering people went with Matlab, while us statisticians went with R. There's plenty of information on Google about why that divide exists.
Julia is more direct comparison to Matlab, and the only reason I can offer why people would not use it versus Matlab is age and wealth of libraries. Julia's a very new language, relative to Matlab, so we may see a wider adoption in the coming years from the Julia project.
Try downloading the AbraVibe collection of functions at http://bcs.wiley.com/he-bcs/Books?action=index&bcsId=632... and running them on Octave. Some of those functions require modification before they will run, especially functions generating graphical output.
EDIT: Also I work with Python, and prefer it to Matlab, although not much for numerical stuff, the majority of department are civil and mechanical engineers who are not that computer savvy as electrical/computer engineers are, they generally know some C and Matlab.