A discussion on researchgate (http://www.researchgate.net/post/Bonferroni-how_is_the_family_of_hypotheses_defined) provided a list of papers, which might help collecting opinions - the papers actually start from the question "when to apply corrections in a multiple testing situation". The papers -all cited often - are:
1) Rothman KJ. No adjustments are needed for multiple comparisons. Epidemiology.1990;1(1):43-6. http://psg-mac43.ucsf.edu/ticr/syllabus/courses/9/2003/02/27/Lecture/readings/Rothman.pdf
2) Perneger TV. What´s wrong with Bonferroni adjustments. BMJ. 1998;316(7139):1236-8.http://static.sdu.dk/mediafiles/D/1/F/%7BD1F06030-8FA7-4EE2-BB7D-60D683B18EAA%7DWhat_s-wrong%20_with_Bonferroni_adjustments.BMJ.1998.pdf
3) Bender R, Lange S. Adjusting for multiple testing- when and how? J Clin Epidemiol. 2001;54:343-9. http://www.rbsd.de/PDF/multiple.pdf
Summary:
1) and 2) focus on "all null hypotheses are true", called the general null hypothesis. It can be more properly rejected (i.e. no alpha-cummulation) if adjustments for multiple comparisons are applied. However, both 1) and 2) oppose, that the general null hypothesis is rarely fully used in the process of scientific research - so the "whole theory breaks" criterion does not automatically apply, when one/some of the null hypotheses in one's data analysis are rejected by chance. 1) adds, that it is naive to think of single null hypotheses, which were (falsely) rejected will never be revisited by the scientific community again.
3) states that once single hypotheses melt in one argument, the adjustments must be done.
From my point of view 1), 2), 3) together just mirror, how carefully we must the "whole theory breaks" criterion. Neither is there a way to just put all null hypotheses in one big sausage - nor a way to rely on the slices of the sausage presented as many single hypotheses. This is, where empirical work really meets working with theory from the domain under research.