It seems like the kind of issue that won't be resolved before cloth simulation becomes commonplace. I look forward to the day it does. That and fluid simulation: I'd like people crossing their legs to result in their thigh squashing a little instead of just phasing through.
This is a bit more complicated than that.
Most games wouldn't want to do cloth sim for the sleeves, because 1. it'd be expensive, and problematic at the point where the soft tissues are pushing against each other and get squashed (and thus the cloth would have to collide in an area where there's little room for it, so lotsa errors) 2. the cheaper trick solutions would usually cover that part better anyway.
The cheap tricks are stuff that we also use on naked arms/legs and the collision models for cloth sims: either an extra joint, driven by the elbow/knee rotation, that pushes out the bony areas; or a hand sculpted blendshape that's also activated by the joint rotation.
In games, the limiting factor would have to be memory and maybe processing power; and of course artist time. Consider that just the hero characters may easily have 3-10 costume variations (although at least you only have to load one per level), and enemy variety is also getting more and more important.
Oh and also, knees are rarely seen and even in those cases, the viewer tends to focus on the face and the hands of the character. The reasonable approach is to focus hw resources and artist time on those areas.
Case in point, just look at the knuckles and wrists of Ellie in the Last of Us 2 trailer, they're practically perfect. I wouldn't be surprised to learn that this feature is only active in cinematics, though.