Hypatia, looks quite interesting. I'm looking forward to seeing more specifics when you're happy with how your app has been crafted. :)
(Although I'm a little unclear on the graph and axis stuff...!)
My only critique of the test is that given the numbers I've run on the average 264 tank, it seems like you may have created an edge case where Stamina would have a very distinct advantage.
As far as I can tell, in a worst-case string of high-end attacks combined with the correct offset of heals would result in a near 100% death scenario: with 2 primary hits, 1 large periodic, and 1 small hits within the same period as 2-3 periodic small heals and one average-sized heal. In this edge case, the margin of death is actually fairly small, and even minor amounts of Stamina would cause the tank to survive until the next large heal internal.
So, at first glance, it would seem as if this test would favor Stamina naturally. This is not to say it is an invalid test (because such scenarios do happen) but it may be a scenario where the end result is somewhat predetermined.
If looking to simulate something like this, I would consider (and may actually spend some time, now that you've put the idea out there!) modeling a more normal boss scenario. In general, the parameters of the boss incoming damage are fairly constant, and the chance of large-scale alignment of hits is fairly uncommon--Blizzard usually purposfully removes such situations, e.g. Gormok and Thorim adjustments.
Instead, I would say the typical cause of tank death is large sources of damage or even normal damage where healers make mistakes or get interrupted.
So, I would actually propose a simulation where the incoming damage does not predetermine death simply due to a worst-case burst scenario--but a scenario instead where typical healing and interruption patterns were roughly modeled. Presuming the tank already has enough Stamina to survive decently (as that is at least the minimum criteria if you know what you are doing) provided nobody makes any major errors.
e.g. Paladin with average size heals using Holy Light spam with average crit rate percentage. Throw in some basic modeling for Divine Plea and refreshing of Jugements, and perhaps a random 'don't stand in bad stuff' movement interruption at a maximum of every X seconds for Y seconds.
While this technically may result in a similar result, I don't think it would be quite so skewed towards an edge case where having only a small amount more health would significantly decrease the chance of worst-case alignment issues. (That said, I'm only presuming that knowing what you said--but given the damage figures and average ilevel you mentioned, it seems likely that it could be the case.)
I may try to conjure up something over the weekend if I am feeling industrious. :)
(I do find it interesting, though, that even with this potential edge case in the scenario, 9 Stamina socket bonuses seem to cause the tradeoff to roughly equal out for Warriors. That actually doesn't seem amazingly far off from Burst Time style assumptions.)

