Imaging Chip Size

[BJL]

I was not trying to imply that a lens designed for a smaller image circle would automatically be worse - just that in balancing all the design criteria Scott listed, one could run into the same compromise resulting in a "small image circle" lens having performance at the edges of its image circle that is not as good as the central part of a lens designed for the FF sensor.

I am sorry, but I just do not see it; for the same focal length, almost everything is easier with a smaller image circle; what design criteria would push the smller image circle design in a direction that sacrifices image quality near the edge of that smaller image circle? On the contrary, removing the need to have good image quality at larger distances might open up some options, like weird asymetrical lens designs that perform better out to the smaller radius at the cost of problems further out.
    Actually, one thing is potentially harder for the smaller image circle; resolution tends to need to be higher to match the smaller pixels typically used. However this seems to be a matter of across the board improvements in design and manufacturing tolerances, with not much reason to think that this would hurt towards the corners.
   In fact, resolution needs are a good reason to think that lenses designed for the job (smaller image circle) are likely to perform better than ones originally designed for a diferent purpose. Canon and many lens makers measure lens sharpness through MTF curves at 30 lp/mm, and probably their designs and quality control are oriented to good performance at upto 30lp/mm. For smaller sensors and smaller pixels, higher resolutions are important, and thus Olympus works with MTF 60lp/mm for its E system lenses. The E lenses' MTF curves at 60lp/mm are close to Canon L lenses at 30lp/mm. (As an aside, according to tests at the PhotoDo site, even very good medium format lenses tend to have somewhat worse MTF at 40lp/mm than 35mm format lenses of the same focal length, but the difference is not enough to offset the extra detail given by the larger image size.)

   On a 10D with its intermediate sensor and pixel size, the comparison would be MTF at about 50lp/mm. How good are Canon lenses at that level? Do their lens designs and manufacturing tolerances ensure good performance at 50lp/mm? Maybe they do great, but in this respect, I feel more reassured with a product that has been designed and tested to work well at the task I am using it for.

[BJL]

However, it's reasonable to suppose that such lenses are still going to be very, very expensive. Most people will have to settle for less and the cheaper 4/3rds format lenses are still going to suffer from a worse performance at the edges and at full aperture ... not just worse performance in relation to the the same lens' performance at the centre, but worse performance compared with a comparably priced 35mm lens as measured 12mm from its centre.

... The Sigma 14mm prime was considered to be quite good with the D30, but unacceptable on a 1Ds.

Comparing the same lens, or lenses of the same focal length, with different sensor sizes is not the relevant comparision; or at least, not the one I am making. What I care about is comparing lenses that give me the same angular field of view. For example, comparing the Sigma 14mm on the D30 to a 22mm lens on the 1Ds, or the 14mm on the 1DS to a 9mm on the 10D. For the reasons stated earlier, I expect the angular field of view (after any "cropping") to be the principal predictor of optical quality in the corners, independent of format size, with the pain level going up as one moves into wide angle territory.

   Also, I do not know about "very very expensive", certainly relative to the alternative of using a 35mm format wide angle lens instead. Consider these "DSLR standard zoom lens" options, all apparently of comparable quality:
$500 for the Olympus 14-54 f/2.8-f/4 [equivalent to 26.5-103 in 35mm based on frame diagonal]
$800 for the Canon 17-40 f/4 [about 27-64 35mm equiv.]
$1300 for the Canon 16-35 f/2.8 [about 26-46 35mm equiv.]
Or for a very roughly equivalent lens for a 35mm camera,
$800 for the Tamron 28-105 f/2.8.

   Using less of that expensive high quality glass does give a cost advantage to making a lens of the same field of view in smaller format. Consider the cost of the quite sharp 7-50mm f/2-2.4 zoom on the Sony 828 2/3" format digicam; its mythical equivalent 28-200 f/2-2.4 for 35mm format would cost more than that whole camera.

   Olympus has given vague details of some lenses planned for next year, including something like a 15-40 [about 28-76 equiv.], probably a cheaper option to go with the second cheaper body; I wonder how that new "standard zoom" option will be priced compared to the Canon 17-40, given their more similar field of view range.

[BJL]

The comparison of performance of lenses of similar field of view for different formats (like 80mm for 35mm vs 50mm for Canon 1.6x format) is difficult; I will take a while to digest Ray's comments.
   However, let me try to make a much simpler point, which arises when seeking a lens for a smaller DSLR format of a certain focal length (or focal length range) and at a certain price: would you would expect better performance from (a) a lens specifically designed for that format and its image circle diameter, or ( a lens designed for use with 35mm film cameras?
   It seems to me that there is a simple argument in favour of option (a) being at least as good as ( and probably better. The designers of the smaller image circle lens have the option of using a design suited to 35mm film format usage; indeed, they probably have such designs available from the olden days when people were still designing lenses for 35mm film cameras. So either they can come up with another design, sellable at the same target price, that works better, or they cannot, in which case they just re-label the 35mm film design as "designed for digital". To make their bosses think they are doing their job, they could probably modify the 35mm design by reducing the opening of the lens's hood, front entrance and other parts, to reduce the angle of light admitted to the lens (since its angular field of view is smaller with the smaller sensor), reducing flare and so improving perfomance a bit, at no significant added cost.
   So, short of sheer stupidity, there is no reason why they should even come up with a lens that costs as much as a 35mm film lens design but works less well, and some reason to think that they will be able to make it work at least slightly better.