Torborn Sjostrand (2015-10-22):

...
But indeed you are right, events with multiple junctions can give
problems. I see that the events that survive have up to three
junctions. (Possibly the ones that fail have even more?) Currently
we have no fix for such complicated situations. So the answer is no,
unless you can substitute with a simpler a "good enough" colour
topology for these events, Pythia is not the way to go for you.
É

Torborn Sjostrand (2015-10-23):

...
By the way, I forgot that also beam remnants can give rise to a junction
by the MPI activity. You can check whether this is the problem by
switching off MPI, PartonLevel:MPI = off. This gives too clean
underlying events, so is not realistic, but may help isolate the
problem. There is a little-used option - for good reasons -
BeamRemnants:allowJunction = off that allows you to keep MPI on but
rejects and tries again anytime the remnants contain a junction.
This might be a messy but possible way to do something. Unfortunately
I am fully committed the next few days so won't have any chance to
look further, but you could have a try yourself.
É