Just ran into that and it was a bear to figure out.
What it was was that I had a static variable (actually a singleton class) defined. Evidently, SSIS does NOT re-initialize a program on second and subsequent invocation, but holds the image and simply re-launches at its entry point.
My Singelton class (and I've verified for several static variables now) does NOT get re-initialized. It still exists. The issue was that it was created with the Dts Variable set that existed on first invocation of the script. Since it's "self" values was not null it never re-instantiated.
When I recognized what was happening, it was of course easy to fix, but one gets used to a stand-alone environment where every program instance has its static values null or set with a static initial value. We presume automatically that a new "run" of the program will likewise have its global spaces "clean" .... in point of fact I'm fairly sure that was what I read as part of the C# "contract" that I'd never need to worry about historical cruft in memory spaces for variables.
Well it turns out that that "contract" is about as good as any Microsoft will make you sign.
It's actually a mixed blessing. Knowing that it happens I can use it to save a lot of overhead in scripts invoked in loops ... but as it's not well, or perhaps un- documented I'll need to be careful to have work-arounds and default loading tests if it turns out not to be true in some future release or version.
(Be gentle in your criticism... I'm new to SSIS. Not so new to program paradigms. CICS mainframe programs would re-init global spaces unless you did things in the linkage to signal it not to ... if you're going to re-invent wheels at least look at old wheels).
-- TWZ