The Mitsubishi/DSM W5M33 AWD transmission is a surprisingly tough manual transmission.  Factory rated for 280ft*lbs, they are known to "survive" 2 to 3 times that amount. Fairly often we are asked, "How much will a Stage XXX transmission hold?" or "Who can build me a trans to hold  XXX power?"  First lets talk about what the factory rating means.

From the factory Mitsubishi rated the W5M33 to hold 280 ft*lbs.  That seems like a low number, and that is because this is an 'endurance limit.'  Meaning a stock trans in good shape will transmit that much torque pretty much forever.    I'm sure this number was a design target, and is padded a bit to help with warranty claims, and manufacturing tolerances.  What this means is I would expect the gears to have an almost infinite life up to around 300-350ft*lbs.

The endurance limit comes from the metals "fatigue limit."  As a gear tooth comes into and out of mesh with another tooth, the load (tooth bending stress) builds up and then releases - this is seen as tensile stress.  When metal is subjected to a variable tensile stress it fatigues - eventually small cracks form, which then grow until complete failure (tooth loss) occurs.  Luckily for us, most steels have a "fatigue limit" where fatigue does not accrue for stresses below this level.  This fatigue limit is approximately 75% of yield strength for single direction bending (normal gears).  For gears used as an idler, the teeth see full reverse bending, and fatigue limit is thought to be as low as 50% of  yield strength.  This is a fact of life, and the tooth bending stress is a function of tooth profile, tooth count, pressure angle, torque input.  The fatigue limit is a function of yield strength, which is determined by the steel type.  For most production carburizing gear steels yield strength is around 100ksi, stepping up to something like 9310 can get you to 200ksi.  In the end, the strength comes down to the material, and the tooth profile/contact.  Sure there are some special tricks here and there to improve strength through improvement of stress flow, and tooth contact, but these will be fractional improvements at best.

You can learn more about fatigue strength from good ole wikipedia https://en.wikipedia.org/wiki/Fatigue_(material)

That being said, a number of tests have been done on the W5M33 transmissions. TRE has been nice enough to publish data showing the W5M33 has an ultimate torque capacity of around 1000ft*lbs.  Through my testing, I've seen that at 550ft*lbs an unknown used gear set seemed to last somewhere around 200 passes at the drag strip.    That being said, we can start to answer questions like "How much will this hold?"  Unfortunately, we can only answer with the question, "How long does it need to last?"   Yes, an oem gear-set will hold 700ft*lbs, but for only 20 or so passes.  If you spend money on all the improvements, like REMISF, shotpeen, detailing, ect, they might last 30 passes.    Let's revisit fatigue for a minute,  and talk about what happens when you hit a trans with a big torque spike - 1 blast of 700ft*lbs can do the same damage as 50 passes at 500ft*lbs.

A lot of times, shops rate the stage 3,4 and 5 transmissions higher than the lower stages, and this is sort of correct.  What you are really doing is buying fatigue life.  Say you send in a core for a stage 2 build that has been on a 400ft*lbs setup for 500 passes - It's half used up.   it will last say 10 passes at 700ft*lbs.    Now say instead you get a stage 3 that has an all new gear-set, your getting a trans that isn't half used up, and will last the full 20 passes at 700ft*lbs.   You are not buying increased torque capacity, you are buying increased life.

The same goes for buying metal treatments such as shotpeeing/REM-ISF/detailing.  These do not improve the strength of the material, they improve the fatigue life.  You are buying life, not strength.

Now lets talk about some of the unfixable design weaknesses of the W5M33.

1. Undercut 4th gear hub - 4th gear is severely undercut at the hub, severely reducing rim strength.  Somewhere over 700ft*lbs, the teeth can be ripped clean off the gear.
2. 3rd intermediate gear used as an idler - This severely cuts down the fatigue strength of this gear.
3. 3rd input gear is 2/3 width of the intermediate gear - This concentrates load into a smaller area sharply increasing the stress and reducing torque capacity.

With all that said, what can we do to improve the life in general street use:

1. DO NOT RUN MORE CLUTCH THAN YOU NEED.  I cannon stress this enough, and I recommend limiting yourself to a clutch that holds around 500ft*lbs.   Say your setup makes 500ft*lbs, but you put a 700ft*lbs clutch in the car.  Every time you launch, or bang off a good NLTS, you are hammering 700ft*lbs into the trans.  Even though it's  small spike, it is STILL EATING UP FATIGUE LIFE!
2. Run a larger turbo, and make power with RPM.  Over 500ft*lbs life goes down fast.  Running a bigger turbo will limit the big torque spike when boost hits, and let you make power higher into the RPM range with less torque.
3. Avoid making pulls from low RPM through peak torque.  It's fun and we all love doing it, but that torque peak eats up a lot of fatigue life.
4. Tune for a FLAT torque curve.

I hope this sheds some light on this subject, and helps you all keep the shiny bits inside the case.