If you LERP (with normalization), your speed will accelerate towards the middle of the rotation and decelerate close to the final orientation again. That kind of "ease in, ease out" may not at all be bad, it could even be desirable.
However, it will still not necessarily work correctly or as intended (and neither will SLERP) if the player is moving.
Imagine you have two quaternions and you SLERP having t go from 0 to 1 in, say, 100 steps. You somehow calculated (that's daunting already!) that you need exactly 100 steps to rotate at the desired speed. All is nice until frame #50 when the player moves.
Now what!
You can finish your old rotation and then start a new rotation to the then-current position of the player. Possibly you will now only need 81 steps (or possibly 140) to attain the desired constant speed. That will look correct, but it is probably not what you want.
You could simply replace the target quaternion. Could you? If player moves in an inopportune direction fast enough, this will give a very unpleasing visual jump. Also, the arc length is now different, so your number of interpolation steps is wrong and you will rotate at a different speed. Bad!
You can recalculate a new rotation from your now-current orientation to the player's updated position, and start SLERPing from 0 again. This is the correct approach, but your movements will still have a somewhat "robotic" feel with a sharp knack in the middle.
If that is undesirable, I would try to solve the problem by attaching an "empty" (no idea what better name to give to it, but e.g. Blender has "empty" objects) to the player with some kind of soft constraint, such as a damped spring with constant speed, and then simply rotate to LookAt(emptyObject) every frame. You will only ever try to rotate directly towards the "empty" (without interpolating N steps), which the player drags behind him as if on a rubber band. Eventually, if the player stops moving, the attached "empty" will catch up with his position, and you will look at the player. No acceleration/deceleration, no jerks, no robotic knacks, no need to calculate how many SLERP steps you need.