If I may -- I would offer a few things to consider:
1. Absolutely correct -- push-pull output transformers make for lousy single ended output transformers, as they lack the air gap that single ended outputs use to maintain an acceptable level of inductance when the core becomes (necessarily) biased if choke coupling is not used. In this regard however, the Mac output transformer is no better or worse than any other high quality push-pull output transformer design.
2. Regarding the operation of the screen grid and the Unity Couple Output Stage, it is true that the screen grid of each tube is cross-coupled to the plate of the opposing tube. However, this doesn't cause the screen grid of either tube to operate any differently in potential, relative to the other tube elements it operates with, than it does in a conventional push-pull arrangement. As a basic point of understanding then, in ANY pentode connected output stage (which includes the Unity Coupled Output Stage), the screen grid of the tube will ALWAYS operate at a much higher potential than the plate does on positive going control grid signals. This is inherent with pentode operation, and why the pentode connection always delivers the most power from a pentode tube, over any other type of output configuration.
In the Unity Coupled Output Stage, it certainly "appears" that the plate of (say) tube A would drive the screen grid of tube B to a very high level (and visa versa) -- and it does. But at the same time, the cathode of tube B is also being pulled by tube B to a very high level, because its controls grid is being driven positive. Since the cathode and plate windings are unity coupled, and the plates are cross-coupled to the screens, that means that the cathode and screen grid of tube B are rising (or falling) at exactly the same amount, so that at any point between quiescent and full power output conditions, the cathode/screen potential remains constant. This represents an ideal form of pentode operation.
Additionally however, while the cathode and screen grid levels are rising in tube B, its plate voltage is falling -- because again, the control grid of tube B is being driven positive. Therefore, the potential between the plate and cathode has now swung very low, while the screen grid remains at the same potential above the cathode as previously established. This then is no different than a traditional push-pull pentode fixed bias arrangement, where the cathode is held at ground level, the screen grid held at a (hopefully) fixed level above ground, and the plate is allowed to swing very low when the control grid is driven positive. From this then, it will be seen that the real issue that the tubes have to deal with in a Unity Coupled Output stage, is the large heater to cathode voltage that exists under conditions of high power output. With half of each tube's total primary winding signal appearing at it's cathode, and the heater circuit having one side tied to ground, this potential can approach 200 peak volts between these elements under conditions of maximum power output, which the heater/cathode insulation within the tube must be able to withstand.
3. So just how hard is the the Unity Coupled Output Stage on the screen grid of the tubes used in the MC225, 240, and 275 designs? It turns out, not as hard as in conventional pentode push-pull designs.
To a significant degree, in Class AB output stages, it is the loading conditions at the plate of the output tube that determine the amount of heat that the screen grid will dissipate under full power conditions. Numerically higher loads allow the plate voltage to swing lower. If the load becomes numerically too high (as in the load line slipping below the knee of the plate current curve), it makes the screen grid -- at its much higher potential -- a much more attractive element for the electrons to move towards, and greatly elevating it's dissipation level in the process. Conversely, numerically lower loads prevent the plate voltage from falling too low, maintaining the plate as the major target for the electron stream to move towards, and lowering screen grid dissipation accordingly.
Using the MC225 as an example, conventional push-pull pentode operation of 7591 tubes has them typically operating into a 6.6K load plate-to-plate, using fixed bias, with a B+ supply of 450 volts for the plate, and 400 volts for the screen. For this tube, these operating conditions produce a high level of power output and low distortion, but operates the screen grids right at their maximum dissipation rating for speech and music at maximum power output. Countless thousands of Fisher and Scott amplifiers and receivers use these operating conditions in the traditionally designed output stages of their units, advertising 35 watts of power output per channel. But McIntosh deviated from these conditions in the 225.
Mac operates both the plate and screen elements from 400 volts and uses fixed bias, but (in part) because of the 100% feedback factor that exists within the Unity Coupled Output Stage, is able to move the load line to the lowest practical value of about 4800 ohms. This raises distortion significantly, but because of the high feedback factor, that is a moot point. More importantly, as a lower (than typical) numerical load, it lowers the screen grid dissipation at maximum power output to levels lower than those produced in all those Fisher and Scott designs using traditional push-pull configurations, while producing (basically) the same level of power output, but at lower overall distortion. Peak plate current is elevated at maximum power output using this load line, but for typical residential power needs from a unit of this power level, that fact will be lost over time to become rather insignificant.
When these facts regarding the impact of the Unity Coupled Output Stage on the screen grids of the tubes used are then coupled with the reduced quiescent current draw (made possible by the 100% feedback factor), it can be seen then that other than the stout insulation required between the heater and cathode elements of the output tube, the design is really much easier on the tubes than most traditional designs are -- as proven out by the historically documented long lives that most quality tubes live out in these designs. In short then, as long as the screen stopper resistors are in place in a 225, any catastrophic tube events in the Unity Coupled Output Stage will most likely be a tube related event, rather than driven by any inherent hardship that the circuit places on the tubes.
As I have commented to Ivan and he has also commented here, we are about down to the crumbs on the original American 7591(A) pieces, with modern production tubes getting notably better over time, but still lacking the quality control that the best American pieces brought to the table. The 7591 class of tubes are very tough to manufacture well, with tolerances no greater than the thickness of a hair on (hopefully) most of our heads to maintain if target specifications are to be achieved. I would again suggest that along with using only the best vendors to cull out the obvious dead beats; one who will burn the tubes in over time to allow the characteristics of those tubes making the grade to stabilize before matching them up, that a visual inspection of the tubes under full power output conditions will then give the most meaningful indication of how well a tube will last over time. Tubes that notably protest under full power conditions (where normally they should not), will still likely be headed to an early grave even if operated at low power levels, due to the invisible, slow torture of the screen grid operating continuously at elevated temperatures. This then gives rise to the "mysterious" failures that occur after just a few hundred or so hours of use -- a failure that full power testing as part of the final vetting process could likely have prevented.
Dave