I think you're entering uncharted territory. I am obviously not a lawyer, but the way I look at it is this:
* The files, unmodified, are the copyrighted property of the original author except where they're a statement of fact. Hence my thread of a few days ago about separating definition and implementation.
* Leaving aside portions which can't be copyrighted due to being a statement of fact, a symbol-by-symbol transcript which preserves the "artistic presentation"- i.e. comments, variable names, and possibly layout- can reasonably be expected to infringe on the copyright of the original author so can reasonably be expected to be subject to his chosen license.
* A cleanroom reimplementation, which relies on an analysis of the original author's code but is provably not a symbol-by-symbol transcript, reasonably certainly does not infringe on his copyright but might infringe on (attempts at) patenting "business methods" etc.
I think that most of us peruse projects on Sourceforge and Github, and use what we've learned as part of our accumulated experience. My own preference is to make "with reference to" attributions in my code so that if it's ever published I can't be accused of evading my responsibilities, and I think that I'd argue that it's no worse than designing equipment or writing a monograph with reference to books and papers in a university library (a company's technical library, which may describe techniques which may only be used under licence, is a different matter).
Finally, I put something on Github a couple of years ago which was a distillation of various material of traceable provenance, which if the rules had been strictly followed couldn't have existed (there was a loop between Torvalds who favours GPL2, some material he'd referred to which was explicitly GPL3, and another party who claimed one of the CCs).
MarkMLl