As a un-opinionated non MF owner I have a few observations.
From sampling theory we see that in order to resolve a periodic function, one must sample at a frequency at least twice the frequency of the sampled data.
In intuitive terms, then if line pairs were considered, say alternating black and white, than a black pixel followed by a white pixel is the best you can do, and is the highest resolving capability of a given sensor.
Given that basis, and ignoring the de-bayering algorithms that synthesize color information, we can generate the following theoretical resolving power:
Camera sensor pixels max LP LP/mm M8Resfactor MaxPrintfactor
M8 27.0mm 3916 1958 72.5 1.00 1.00
D3 36.0mm 4256 2128 59.1 .82 1.09
P20+ 36.9mm 4080 2040 55.3 .76 1.04
P25+ 48.9mm 5436 2718 55.6 .77 1.39
P30+ 44.2mm 6496 3248 73.5 1.01 1.66
P45+ 48.9mm 7216 3608 73.8 1.02 1.84
Neglecting for a moment anti-moire filters and other factors such as noise and lens capabilities we see that these various cameras have a fairly narrow range of resolving capability but a larger range of print size capability at the same on-sensor resolution. That means that given equal capability resolving power lenses the P45+ can produce a print that has linear dimensions 84% larger than the M8 with about equivalent on-print resolution. Further, I see no significant advantage to using the P20+ back at all, I'd rather carry an M8.
Next let's think about noise.
We often pixel peep and complain about noise that gets lost in printer dithering, but that being said, there are two factors that affect noise. One is the stochastic nature of photon distribution over the sensor and the number of them that must strike a pixel to raise it to a given charge threshold. The smaller the pixel size, the larger this stochastic noise will be. It is just the nature of light in the particle view of things (please don't get me started on the wave issues). This becomes worse as light levels get lower and at the theoretical lower limit at true DMax-limit there might be an equal probability of a photon strike on a pixel or not at a 50% noise threshold during the exposure duration. Of course that photon strike may or may not knock one electron's charge which is as low as we can expect to get and that is a function of lattice doping and other lossy factors such as passivation, reflection, metalization, and filter absorption. Being as lazy as I am, I am not going to calculate this, but I estimate that this loss amounts to at least an order of magnitude, but I digress. The bottom line is that for a given sensor technology, the ratio of pixel area determines the ratio of threshold dark sensitivity. In addition, CCD and transistor based sensor cells both suffer from a saturation phenomenon which roughly means that the larger the cell's area, the greater the number of quantizations of charge that may be theoretically measured. So, twice the area and twice the potential iso as well as twice the potential dynamic range.
The M8 employs a 6.8 micron square pixel and shares this size with the P30+ and the P45+. The P20+ and P25+ are both 9 micron and the D3 is between at 8.45 micron.
From this, given equivalent analog to digital conversion, I would assume that noise levels of the M8, P45+ and P30+ could be equivalent, the D3 would be lower, and the P20+ and P25+ lower still.
How this translates into iso levels is a bit dependent on the quality of the a to d conversion as well as the manufacturer's decision on how much black level noise is acceptable in their nominal iso rating. What it means to us is that a large pixel sensor will yield lower noise and smoother gradations in the darks.
So, for my money I have decided that:
For low light hand held stuff I got a D3
For portable use with decent performance I am keeping my M8
When I want larger prints (and I just might) it makes no sense for me to buy anything less than the P25+ unless I am willing to invest in MF lenses with M8 equivalent lens quality which would indicate the P45+ or maybe I just want the MF way of doing things.
-bob