I once worked on a "barely working" application - the original project (which I wasn't involved with) had clearly been a train wreck and it had delivered a tiny proportion of what had originally been promised.
It presented the user with a couple of fairly straightforward screens, did some integration with a couple of other systems but had over 6,000 (not a typo) classes and a dependency on just about every acronym in J2EE. This had apparently taken a dev team of 30 or so over a year.
It was totally replaced by a much smaller application written in a couple of months by a team of three that provided a sound basis for delivering much more functionality than was originally expected.
So sometimes, rarely, chucking away something that barely works is very much the right thing to do.
"If you can’t implement new features in a clean way I claim: You aren’t really able to implement the whole application in a clean way from scratch either"
is noble, there are surely circumstances where the existing system is so broken/incompatible that this claim might not hold up. However, you would hope competent developers are aware that outright binning a working system is a Bad Thing.