The first ones needs to be fluent in every types of algorithms and data structure. The other type will have the task of modeling real world processes into a computer (which means "concept organization translation" rather than pure algorithms).
The first ones deal with primitive types, and very rarely more than that. The second ones need to model all the entities of a real life problem and their complex interactions (complex in the sense of combinations and organization).
That's why I think there will always be a gap between those two developers. They are dealing with very different kind of tasks, yet their jobs titles are the same.
Actual the second group is occasionally saying, "WTF was the first group thinking?" after spending hours trying to get the tools to work for the real world. (Try spending a few hours/days writing wrappers for PKI software with shitty APIs...)
The distinction is real, in that many programmers never do build basic tools, while it is artificial in the sense that there is no reason for this other than lack of skill and organizational culture that says that that is not something you are supposed to do.
Where exactly is the distinction? How would you classify Monodevelop, or git/svn, or the C++ STL, or SQLite, or Emacs, or the Unix shell utilities, or Photoshop?
http://steve-yegge.blogspot.com/2007/06/rich-programmer-food...
"If you don't know how compilers work, then you don't know how computers work."
Becoming better at the first class of programming that you describe, will also make you better at the second class.
And a lot of times, "modeling the world" is a pitfall of object-oriented programming, and something that you don't really want to do. But it's very tempting if you've only ever programmed in Java.
IMO this helps at the most very basic level because it lets us consider "hey, I wonder if there's an algorithm out there that'd help me with this" and gives you the impetus you need to go looking.
Kind of a "you know what you don't know" situation of sorts.
For example, I'm writing a pure JS database (https://github.com/louischatriot/nedb), and using indexing I was able to speed it up hundreds fold. That's clearly not something I could have done without knowing about self-balancing binary search trees.
I also think that knowing algorithms does make you a better programmer, even if you don't use them on a daily basis.