Even the experts admit that it's untested, and every law student knows that anything up to an actual test in an actual court is little better than a guess. Also, beware of "experts" who aren't even talking about copyright law. For example, consider the three cases mentioned in the "US" section here.
http://en.swpat.org/wiki/Implicit_patent_licence
The Hewlett Packard quote is clearly about products, which I know for sure are not legally the same as licenses. The Bottom Line quote's mention of purchases casts a similar shadow. Only the De Forest quote passes even a rudimentary sniff test. The case for this implicit license is weak indeed.
But fine, let's say there is such an implicit license. I have good news for you: it would apply to the Facebook code as well. The same principle used to argue for the existence of an implicit patent-license grant under plain BSD - i.e. that trying to bring a patent action would interfere with the still-operative copyright permission given in the LICENSE file - would still apply. Nothing in the PATENTS file would change that.
Belt and suspenders still beats belt alone, no matter how strong or weak the "implicit license" belt might be.