The maximum possible total amount of light (exposure) depends on both the sensor surface area and the maximum lens surface area. Increasing exposure increases S/N ratio. Perceived image quality is highly dependent on S/N.
Fast lenses have more surface area. Any disadvantage of smaller sensor surface areas is negated by increasing lens surface areas.
This has nothing to with marketing and everything to do with versatility.
Very true and very rational. And of course a further point to be made is that designing super fast lenses is easier when dealing with small format sensors given that the outer area of the image circle is "thrown away" and not captured. So there are now more opportunities being exploited by lens makers.
But in general, I am convinced that perhaps 80% of the desire for fast lenses is purely personal and aesthetic and driven by the market. People these day like the idea of nice out of focus areas - good bokeh - in lenses that can be described as "character" lenses. Anything that hints that it
might be able to offer this gives it a marketing edge.
Even if it can't - and many fast lenses seemingly can't really at least not without significant compromises. Yes they are fast, in some cases very fast, but someone recently pointed out that there seems to be an inverse relationship between sharpness wide open and good bokeh in OOF areas (at least in normally priced lenses - lets leave the Leica Noctilux out of this for the moment).
I don't think this is strictly a universal rule but there does seem to be truth in it - firstly it is just more difficult to correct a fast lens even with the benefit of a small sensor size and hence fewer constraints. Secondly, a common complaint about fast lenses also seems to be that either that they are sharp in the centre but not beyond. Or that they have good sharpness across the field, but have harsh bokeh. Getting both is just plain hard and expensive. getting both in a very wide angle lens - tougher. etc.
But of course today we have great computers that can run millions of ray calculations in complex lenses per second so designing "super" lenses is more possible. What is hard, I submit, is how to design for nice bokeh as this is essentially an aesthetic choice which would be difficult to design for purely using computers - I imagine you have to design it, then build a prototype and test it using the mark one human eyeball to look for its image characteristics. Then perhaps redesign and re prototype it if the results are not what you want. Hence it costs much more. So instead we end up with fast lenses that may or may not also be good character lenses that people like.
(Not really your point I know but just my pennies worth.)
