[snip]
If I had to choose between very fragmented or completely uniform, however, I'd choose fragmented.
Hm, I would prefer less extreme options, but I certainly agree that some fragmentation is to be preferred to a very uniform Linux. I mean, I like systemd and I do think the systemd group have some very good ideas, but they certainly don't include e.g. KDE in that future as anything else than token gestures. I would be very unhappy if Gnome ruled the Linux desktop without real competition. I have no animosity against Gnome, I just love KDE.
So there...
We can't predict where Linux will be used in the future, and so we may need the core-level diversity that fragmentation brings. It's about more than just where libraries are placed, but about ways of doing things. Being able to drop in an alternative system-level structure lets us try out new principles, such as systemd versus sysvinit for instance. We might all be using systemd in 10 years, but I would bet you nobody will be in 50, so if we're no longer able to experiment with alternatives because we're locked into one system, that new alternative will come from outside the Linux ecosystem. It's evolution: stop growing and settle into a niche, and eventually something nimbler will outcompete you.
I agree with the general gist of what you say; Linux will only be healthy with at least some internal competition of ideas and how to implement them. As it is now, people can move from e.g. Linux KDE to Linux Gnome, if the KDE direction bothers them. With only one DE on Linux, they can only move to other OS's if they don't like what they get.
The same thing also applies to low level libraries, programming languages etc.
But on the other hand, to reduce the fragmentation is also very worthwhile goal, so I would hope that systemd could be compromise between the two extremes. IMHO, I do think systemd so technically superior to all other script based init systems, that it barely has any real technical competition anyway.
Regarding systemd as long term solution, and lock in. In some ways I think systemd have paved the way making future init system shifts much easier; with rather firm external API's it is much easier to gain momentum if the new solution is backwards compatible.
The new direction could either be a fork that later changed in a new direction, or a whole new system, that just provided external compatibility like systemd did. I am not suggesting that especially the latter will be easy, but it is doable, and since almost all distros at that point would be using systemd, they could easily change init system.
It would not be a one man solution, however; organizing and cross distro cooperation would be the key to success
As it is now, I think that a fork of systemd is the most realistic long term solution, if one find it necessary that Linux have competing projects on every level.