For a start, I would be very surprised to hear there is a general strategy to solve the problem which is not so abstract as to be useless.
Now, showing the variation is safe is much, much easier than proving it: your task is to prove the variation is safe? If so, you might minimize the number of parts (reduce the number of philosophers to two or maybe three) and try ta apply one or more formal verification methods.
In practice, formal verification is so rare that I might illustrate it best with an example: I've talked with people working on software used in nuclear facilities (haven't gone into the details) and when I asked if they use formal verification, I got "well, yes, we should [be using it]..."