Allow me to "unpack" the mechanics of Leica's rangefinder focusing a bit.
Let's start with the "fixed" - or at least unchanging from lens to lens - RF mechanism in the camera. Through levers and fulcrums it takes an in-and-out movement of the lens or camera roller, and transforms it into an angular movement of the prism that transmits the secondary rangefinder-patch-only image in the viewfinder. We see that angular movement as a "panning" side-to-side movement, such that we can focus the lens until the two RF images "pan" into alignment. The geometry of the fulcrums and levers does not change, and therefore there is a fixed relationship between the amount of lens movement in and out, and the amount of side-to-side movement indicating a given focusing distance.
Thus (made-up numbers for illustrative purposes) a 1mm movement of the lens forward from infinity will always produce, perhaps, a 2-degree rotation of the prism and movement of the RF image, thus indicating, perhaps, that focus is now at 50 feet/15.15 meters.
HOWEVER - that only applies to a single focal length - longer lenses will move more to get from infinity to 50 feet, and shorter lenses will move less to change focus from infinity to 50 feet. For reasons of historical contingency, the geometry of the Leica rangefinders assume one is always using a focal length of 50mm exactly (the original 50mm Elmar - except not even then).
Therefore, any lens that is not exactly 50mm in true focal length will misfocus, if the roller simply touches the back of the moving optical cell (barrel holding the actual glass). Instead, every lens Leica makes has a cam: a separate brass inner barrel that moves in and out to simulate the movement of an ideal 50mm lens, while the actual lens moves in and out a different amount. Deep in the lens, the cam that touches the RF in the camera is threaded inside the focus ring that moves the glass, with a differential threading that moves the cam 1% or 45% or 270% LESS than the actual glass movement (for 52mm or 90mm or 135mm lenses) and 70% or 42% MORE than the actual glass movement (for a 35mm or 21mm lens).
OK - we are good so far. The actual movement of the cam and rangefinder is now the same regardless of how much the actual focal length being used is moving. "50 feet" looks the same to the RF with a 21 or a 50 or a 90, even though the lenses are actually moving very different amounts.
But the next flaw in the ointment is that a "50mm" lens is not always (or even usually) actually 50mm in focal length. Most Leica "50s" are really something other than 50mm - they are 51mm or 52.2mm or 51.5mm in true focal length. "90s" may be 90mm, or 89.5mm, or 90.5mm, etc. etc. Leica builts a purported 90mm lens - and then tests it to see what it really is - 89.5 or 91mm, or whatever. And then tweaks the cam to reflect the real focal length and real movement needed for correct focusing - and also engraves a little 2-digit number on the focus ring beside the "m"-for-meters engraving, to show the actual measured focal length, for future reference by the repair techs if/when they need to know (and the enjoyment of Leica collectors ). Check your own 50s or 75s or 90s or 135s for the little numbers - "16" on a 50 means it is really a 51.6mm lens; "22" means it is a 52.2mm lens; for 90s, "00" means it really is a true 90mm, and "10" means it is actually a 91.0m lens. Even the Barnack Elmars come in different "flavors" - from 50.1mm to 51.9mm.
BTW - this extra precision is usually only needed for normals and teles - wide-angles have enough depth fo field even at f/1.4 (or other factors such as focus shift that produce much bigger errors anyway) that the focusing geometry works fine regardless of whether a given "35mm" is actually 34.6mm or 35.5mm. So they do not have the "true focal-length" numbers engraved.
HOWEVER (again) - even that correction for true focal length is not always enough to get a high-performance lens into focusing spec at longer focal lengths and larger apertures, and therefore the back surface of the brass focus cam in a lens (which will contact the camera's RF roller) may be machined delicately, one lens at a time, to adjust the in-and-out movement (subtle hills and valleys) as the cam rotates with focusing, nailing focus at various points (with smooth transitions between points) for that specific unit.
But note that that means that if your camera RF roller that contacts the lens is off, to one side or the other, by even 0.1mm - it may be "feeling" a different part of the machined cam surface than Leica's factory-standard test camera did. Thus one may get generally good focusing, and correct focus at infinity using the stop/lens scale, but the RF images may not align perfectly at infinity (which is, after all, not possible to measure exactly anyway - the Leica RF geometry and focusing image will look the same whether you focus on a building 2 miles away, or the moon 250,000 miles away - either one counts as "infinity" to the rangefinder.
Note - the 90 APO-Summicron and the 135 APO-Telyt (and some other older lenses, e.g. early versions of the 75 f/1.4 and 90 f/2 (1980) or 135mm Tele-Elmar) actually have their cam surface way up inside the lens, with a narrow (blackened) spring-loaded brass bar or rail that transfers the in-and-out movement of the cam to the camera - via a tip like this - \___/ - that is all you see sticking out the back. That system saves some weight, and also eliminates the problem of where your roller actually touches the lens feeler (it doesn't rotate side to side, it just moves in and out.) But is extra complexity not used in shorter focal lengths.
At any rate, the short version is there are lots of complexities working behind the scenes, and each complexity can introduce a tiny error here or there at times. It may or may not be enough to produce a noticable focus problem, for any given lens sample or photographer.
Edited by adan, 14 June 2018 - 08:19.