Perhaps I should step in here as a gnuplot developer with a bit of historical perspective. The "weirdness" you see is due to the evolution of the concept of open source and open licensing terms over the past three decades. When you read the gnuplot license, you should be aware that you are reading a document from more than 25 years ago.
Gnuplot has been distributed under this same license for longer than the GPL has been in existence. Gnuplot version 1.0 was released by the original authors in 1986, and a much expanded version 2.0 incorporating contributions from a larger development group came in 1990. Up to this point there was no connection to the GNU project or the FSF. The "gnu" part of "gnuplot" is purely coincidental according to the original authors. Although this was all before my involvement with gnuplot and therefore out of my direct knowledge, my understanding is that starting about that time early versions of gnuplot with this same license were hosted as part of the GNU project even after GPL1 was drafted. If anyone has specific documentation to support or contradict this, I would appreciate having a copy for the project records. To my knowledge, no connection to the GNU project or to the FSF continued past the point where gnuplot moved to being hosted on SourceForge, but again this was before my direct involvement so I am hazy on the exact timeline.
Several years back we (the current gnuplot team) made an effort to contact as many early developers as we could find to see if there was a more modern license that everyone would agree to. As you might imagine, we failed to track down everyone. Furthermore, even among the people we did get responses from there was no such universal agreement that would allow a license switch for the existing body of code.
We did introduce guidelines encouraging dual-licensing for any new self-contained modules or subsystems contributed to gnuplot, and if you look you will see that several of the new terminals are dual-licensed GPL and one is dual-licensed to BSD. This works nicely for individual terminal support because each terminal type (i.e. output mode) is a module that can be configured in or out of the program when building. It doesn't work so well for contributed modifications to the core plotting code.
Last year (2011) a review by Debian identified a single problematic file in gnuplot, dating back to contributed 1980s freeware code whose license restricted commercial use. This file was used only to support dot-matrix printers and similar 1990s vintage output modes. As a result, these devices are not configured or supported by default in gnuplot version 4.6.
Bottom line? There was no intent to choose a "weird" license. Rather there was a very forward-looking intent 25 years ago to release code under what we would now call an open source license. The GPL did not yet exist as a model.
We might have a separate, interesting and perhaps flame-attracting, discussion about how well the GPLv2 or GPLv3 would have meshed with gnuplot's original design intent, but absent a time travel device to provide Thomas Williams and Colin Kelley with an advance-reading copy this is a moot point. Spilt milk under a bridge never crossed.