If we know the position of the pixel and the light, can't we just do this in the lighting pass:
float shadowFactor = pixelDepth < length(LightPosition - pixelPosition) / FarClip;

Where did the value pixelDepth come from? pixelDepth needs to be the scene depth as rendered from the light... i.e. the shadow map.

If we know the position of the pixel and the light, can't we just do this in the lighting pass:

float shadowFactor = pixelDepth < length(LightPosition - pixelPosition) / FarClip;

I'm sure if this worked someone would have figured it out already, but I can't see why it wouldn't and it's bugging me.

There are two possibilities:
1. pixelDepth is the depth from the GBuffer, in which case your formula makes no sense (if the object is further from me than it is from the light, it is shadowed).

2. If pixelDepth is the depth from the light space, then it must have come from a shadow map. Sure, what's confusing you is that you can retrieve the data from the GBuffer and turn it into light space, and obtain "pixelDepth". However in that case your formula would always pass (never shadow). The detail you're omitting is that in the algorithm pixelDepth may not be from the obtained from the object you're evaluating, but rather from another caster that is close to the light (and as pointed out, that caster can be off screen, but even if it isn't, your assumption won't work).