Stoney's answer touches on the key aspect of the phenomenon that is addressed in the question. There is a rule in syntax that is valid for many European languages (and perhaps for other languages as well) that can be formulated as follows:
Constraint (first formulation)
A premodifier of a noun (e.g. attributive adjective) may not itself have a postmodifier.
In slightly different words:
Constraint (second formulation)
A predependent of a noun may not itself have a postdependent.
The examples produced in the question that are awkward all violate this constraint. Here are some further examples:
(1) a. ??a proud of his children father
b. a father proud of his children
(2) a. ??the upset with the problem student
b. the student upset with the problem
(3) a. ??a crazy about politics activist
b. an activist crazy about politics
The a-examples all violate the constraint, whereas when the adjective phrase is positioned after the noun, the structure is fine.
The motivation for the constraint probably has to do with processing efficiency. If one examines the syntax trees for these examples, the a-structures all contain an elbow of a sort that immediately precedes the noun, whereas the b-structures no longer have such an elbow. Syntactic structures that consistently climb (left branching) or consistently fall (right branching) are easier to process than syntactic structures that mix left and right branching. There are numerous phenomena of syntax that are sensitive this aspect of syntactic structures, e.g. extraposition and shifting. These mechanisms serve to reduce center embedding (i.e. the presence of elbows in the structure) and they thus aid processing. The constraint here concerning the pre- and postmodifiers of nouns appears to be a general principle of syntax that also serves to aid processing. It reduces center-embedding (i.e. elbows). This phenomenon is discusssed in detail here:
Osborne, Timothy 2003. The Left Elbow Constraint. Studia Linguistica 57/3, 233-257.