As an inveterate user and observer of open source, I've read many criticisms of open source software development that simply don't hold water. This post is about one criticism that I believe does hold water.
In contrary to the straw man often erected by anti-open source pundits, most open source projects have a relatively formal management structure wherein one person or a committee exercises strong top-down control over the project's canonical source code repository. That dictatorial hand may be felt lightly or heavily, but at the end of the day decisions must be made, and priorities set, and the project will follow the dictator/committee's decisions.
(Or --- in the case of a fork --- not, but nobody likes forks and ultimately each half of the fork generally assembles some sort of management structure.)
However, not all open source projects have a dictator. A few projects manage to survive with only informal technical governance structures. The largest and most successful of these is probably KDE, a desktop environment. KDE has a number of associated formal legal entities responsible for things like legal and marketing issues, but the technical direction of the project is not managed by either a benevolent dictator or a committee.
There are good things and bad things about this. Here is one of the bad things: this KMail wishlist item has been in the bugs database since December 2000. It is the top vote-getter on the most requested features list. But nobody has tackled it. A couple of people have made gestures in the direction of implementing it, but no bug-closing patch has been committed to the trunk.
In an open source project with a dictatorial or committee-led governance structure, somebody would long ago have cracked some heads and gotten this feature implemented. In a commercial software project, open source or non-, some engineer would be assigned ownership of this feature; and goddammit, if that feature didn't get implemented and maintained, that engineer would be fired and the feature would be assigned to someone else. But KDE's headless. It's less like a mammal with a central nervous system than an enormous amoeba whose various pseudopodia ooze tropically in the direction of "developer itches" and "coolest implementation hacks" (hence the recent proliferation of "hugely ambitious infrastructure refactoring" subprojects like Plasma or Solid) rather than unsexy, annoying-to-implement features that merely provide value to end users.
Now, as a by-product of this process, you often end up with a lot of excellent software. I use KDE every single day, and on balance I prefer its interface to both Windows and Mac OS X for intensive use. But this software process isn't rationally optimized to serve the end-user, and so there will always be these frustrating little blind spots.
Which isn't to say that commercial software processes are rationally optimized to serve the end user either, but that's another whole series of posts waiting to be written.
 Yes, that's right --- not the code, not the customizability, not the Free and open source license, but the user interface of KDE is preferable, for a heavy user, to either of the big consumer operating systems. Windows XP and Mac OS X ---especially the latter --- are both highly polished and acceptable for casual use. However, when I'm juggling several terminals, eight or nine Emacs windows, and a half-dozen Firefox windows, the extra features of KDE's window manager simply destroy even Mac OS X's. Ironically, some of KWin's best features were in the classic Mac UI, and were dropped in OS X. (I first got hooked on windowshading when I was an undergrad Silicon Alley web production intern, juggling BBEdit, Photoshop, and multiple versions of Netscape. Ah, the old days...) Other features, like window gravity, have been pretty standard in Unix window managers for years, but have somehow never cracked consumer OSes even though they seem like no-brainers.