Viewshed analysis generally only checks the theoretical line of sight - can a cell be seen from another cell, without any higher terrain standing in the way?
It doesn't take into account whether or not the human eye can make out an object at that distance. It will (if you choose) take refraction into account (the slight bending of light through the atmosphere over large distances) but not visual acuity.
The actual visibility depends on the size of the object and its distance, and the angular size - this site here has a useful calculator.
The formula is fairly straightfoward and can be found here
angle = 57.3 * (size/distance) where distance >> size
For example, a 1m wide telegraph pole 1km away has an angular size of 0.057296 Degrees. A 10m wide building 12km away has a similar angular size of 0.04775
According to wikipedia, the human eye can resolve at a resolution of 0.02 degrees, so you'd be able to make it out, assuming normal vision. According to that site I posted, the telegraph pole would hit that 0.02 degree limit at around 2.8 km.
If you assume the smoke is an opaque column and can estimate its width, you could measure the distance and use that formula.
I think SAGA GIS can produce a distance raster as part of its viewshed output. By using raster calculator and that formula, you could find - for each cell - whether or not the smoke would be visible. That would also tell you the distance you would need to extend your viewshed analysis to.