Oh boy, I sure hate to pour cold water on so much of what has been said so far but, in my experience, here are the facts as I have found them in the real world.
First, a 50hp 1750rpm DC motor has exactly the same continuous torque range as a 50hp 1750rpm AC motor.
Second, as for overload capabilities, a NEMA Design B motor will deliver a max of 220% short time overload if across the AC line. On an inverter, generally the overload capability is much less due to the short time overload ampacity of the INVERTER. If you oversize the inverter compared to the motor, you could and I sometimes do take the motor all the way to the 220% limit. A DC motor is also capable of 300 to 400% short time overloads but, again, the DC DRIVE is the limiting factor since these are generally sized for 10-50% additional ampacity for overload purposes. Further, if you try to overload a DC motor heavily at slow speeds, you will very likely distort the commutator bars and an expensive repair will result. In practical terms, a DC motor quits around 220% just like an AC motor.
Third, as for motor cooling capacity, a TEFC DC motor has very limited low speed cooling just like an AC motor and for the same reason--low fan speed. If the DC motor is in an auxiliary cooled or TENV enclosure, then it will basically cool down to stall. And so will an auxiliary cooled or TENV AC motor!
Fourth, the use of an encoder or analog tach on DC systems has nothing to do with low speed torque. They are there for speed regulation so the speed remains constant regardless of motor load and a few other obscure variables in the drive. While a pulse encoder produces a near perfect signal to follow, the average AC or DC analog tach is good for about 1% error. Your resulting system speed cannot be better than that, can it? The main reason you see so many tachs on DC systems is that, without one, the speed error can go as high as 3-5% which is often unacceptable. With AC, the speed error is primarily a matter of how well the drive compensates for motor slip. Using 4 pole motors for illustration, it used to be common for the nameplate to read 1740-1750rpm. That's 3-3.5% speed error which the drive hopefully can reduce somehow. Today, the same size premium efficient motor will typically be labeled 1770 to 1775rpm. That's only 25-30 rpm (less than 2%) speed error for the drive to manage. I recently started up a 400hp system with a motor labeled 1787rpm!!! That's a better than 1% speed regulator before the drive even tries to help further! My point is simply that the choice of motor is a key component in an AC system's speed accuracy. Now, you don't have to have a closed loop flux vector drive to get excellent torque and speed control at zero and near zero speed anymore. Today, the best sensorless systems can do that job in the process reducing the speed error to around one-tenth of motor slip. Of course, if the speed has to be dead-on or the error has to be non-cumulative as in tensioning applications, then the AC system needs a pulse encoder too.
Finally, let's say you size an AC system at the same hp as the old DC system and you find that more starting torque or higher low speed cooling or less speed error would be desireable. Simply increase the drive train ratio between the motor and the load so, at maximum load speed, the motor is turning at the 90Hz speed instead of at the 60hz speed. All of those benefits come your way with this simple chance. You say the motor is direct coupled and the drive train can't be changed? Then do as suggested above and use a motor of the same hp but with more poles, as a six pole (1200rpm) rather than a four pole (1800rpm). All of those benefits are yours this way too. The only disadvantage with the slower motor is a somewhat more expensive and physically larger motor. But, for the price of a sheave or sprocket, it's cheap insurance if your project is in trouble!