As I have used Pentax since the inception of digital (and 25 years prior to that on film) and had a ton of old glass I had lots of opportunities to try out IBIS. What I find overall is the following, all of which are my opinion based on use and not scientifically measured.
Starting with the K10D over the *istD I found IBIS did provide 2-3 stops improvement, and this was especially true using shorter focal lengths in low light. I found for longer focal lengths, it was not as effective.
Moving to the K7, I started to notice improved performance with tele lenses, and even posted one shot here of a night heron, taken at 1/40th with a K300/4 on the 1.7x AF converter, I will admit, however, this was probably the best shot I ever took, relying totally on shake reduction, as 1/40 with a 500 mm lens on a crop sensor totally defies the 1/(F x crop factor) rule of thumb which would be 1/750 minimum shutter speed, as it is about 5 stops better. I believe several things helped this shot.
- My technique, with my eye against the viewfinder, one hand on the grip and shutter release, one hand on the lens hood, as far away as possible, stance of feet apart and right leg stepped back about 1 foot, and elbows tight on the body I.e. the classic recommended technique
- the lens, not that the K300/4 is a great lens, in fact I think it is rather average, but at 1040 grams it is relatively light, and compact, especially when combined with the 1.7x converter. 500mm at 1.2 kilos is REALLY LIGHT and the center of gravity is not that far away from the camera itself
- finally, and obviously there were improvements in IBIS between the K10 and K7.
For a long time after this I fell away from relying on IBIS because my main lens for birding was a sigma APO 70-200/2.8 EX and Sigma APO2x EX DG Teleconverter and none of the bodies would allow modification for use of a TC to modify the SR focal length. This was, IMO a mistake by Pentax because they didn’t make a TC themselves, Didn’t for years have an offering of their own beyond 300 mm yet would not accommodate users (and customers) who sought 3rd party solutions for failings in their line up.
The other thing I noticed, and this was with the IBIS implemented in the Q, is that it seemed to be over active, especially with longer lenses, and in many instances I found the best stabilization was at about 2/3 the actual focal length. Where I am heading with this observation links into some of the old observations in this thread, where some manufacturers are claiming greater improvements with some lenses over the general capabilities. This is specifically based on how IBIS has to work.
Regardless of who makes it, or how many sensors, in a camera, all the accelerometers are very close to the focal plane. They have to be because a DSLR is , except for mirror box and grip, about 25mm thick, and a mirrorless, or a small format camera like the Q is substantially less. Therefore to calculate how much compensation is needed, and considering that any movement is most likely in the form of a rotation of a free body about its center of mass, the IBIS really needs to know. Where the center of mass is, for each and every lens, and the range of movement during focusing from min focus to infinity. Without this data, IBIS tuning can ever only be approximate, and would likely support shorter focal lengths over longer ones, and lighter lenses over heavier ones. In fact, this last point could probably be tested easily. I have two lens candidates, to propose the test, my tamron 28-75/2.8 and my sigma 70-200/2.8. Using both lenses at 70-75mm and varying amounts of “need” from the shake reduction, could with enough samples demonstrate the pros and cons of each lens, and how the body manages shake reduction.
The other thing I note in a lot of the old discussion is moving to a formula based on sensor resolution. This is perhaps an incorrect approach. You need to go back to the film era, to get to the root of a lot of rules, specifically with respect to the definition of acceptably sharp.
The definition of acceptably sharp, was that a point of light, would be no larger than 1/100 of an inch when a negative was printed on an 8 x 10 paper. This led to the definition of circles of confusion of .03 mm for Full frame and .02 mm for APS C sensors.
Now consider the resolution of the sensor. A K1 for example has a pixel pitch of 4.9 microns and therefore there a 30 micron spot will span a little over 6 pixels doubling the resolution will make the same spot 12 pixels, but it won’t change at all the size of the spot when printed to 8 x 10. So the only way to gain a claimed improvement in SR performance as a function of resolution is to enlarge further and redefine the circle of confusion.