An interesting thread. It appears to this reader that there be two main schools of thought -
(a) those folk who see negative training in pushing the pilot's individual boundaries, especially if the aircraft's systems have to be somewhat prostituted to achieve an end, and
(b) those folk who see some value in stretching things a little for the (sometimes) difficult to quantify potential value to the pilot's overall skill set. There is the ever-present rider that such things must be consistent with a simulator's reasonable capabilities ie, if the simulator's performance is hopelessly unrealistic the exercise ought not to be countenanced at all.
Clearly, if an event is an impossibility (or very nearly so), there is an argument which suggests it is of little, if any, value. It would be expected that folks in the first group might opine in this direction.
Those in the latter group might see some (un)related value if the set up is used as a means to an end to establish a scenario.
To this reader, the essential consideration (given the realities of real world airline flying, training, and checking) is that the reasonably probable needs to be isolated from the highly improbable with a guarantee of "nil hazard" in the latter case so that the pilot might be prepared to expose himself (herself) to the potential for embarrassment, etc. There is the added difficulty in the case of a check pilot holding regulatory delegations in respect of delegation integrity - I've never been too sure how this one ought to be addressed - possibly by removing the delegated check pilot from the overtly extended training environment. The specialist training pilot should not have the same consideration with which to battle at a personal level.
I put myself squarely in the (b) group, with the caveat that a declaration of "nil hazard" is made (along with a rider that, if we collectively screw it up really badly for whatever reason, then we will go back and practice whatever it was until we can achieve an acceptable performance level). I note that I have had some very experienced check pilots in this situation - provided the initial session briefing was sensible, in all cases, the exercise was completed in good humour with a reasonable achievement of the intended manipulative skillset end.
An example I use from time to time relates to a 732 simulator contract with which I was involved many years ago. The particular operator made extensive use of overspeed takeoff schedules (line and sim) but had at least one sector which was flown, routinely, at a very low weight and for low speed schedules. It was clear that there were surprises to be had as the crews had no exposure to this "tiger country" end of the aircraft's handling spectrum. Mainly for the initial command/intake folk (which was the main thrust of our contract) I provided appropriate work up to the point where a min weight, min speed, aft CG, failure could be handled adequately hand flown, raw data, IMC, and with tracking out on the opposite end localiser. There were a couple of other exercises introduced with similar reasoning.
Was this stuff of much direct, routine value to the flying side of things ? Of course not.
Was there value to the development/improvement of stick and rudder I/F work (for that was the underlying intention) ? You bet there was ! (I note that I used a range of skillset exercises as part of an overall work up of basic I/F handling skills which were a bit lacking in some of the upgrade personnel. By the time they had finished my gentle ministrations, the problems were, in the main, adequately addressed).
Of relevance to the thread topic, I used this exercise (only with, and by, agreement with crews), for exposure during after LPC work during available sim play time. Those who wanted a go without any lead up practice saw what constituted "startle" in stark and harsh relief. For the old hands, a couple of practice work up runs was the general route to an acceptable level of stick and rudder proficiency for that box's presentation.