This excellent chart by Bill Claff suggests that after ISO 400 you get very little IQ improvement raising the ISO on an EM5 - ...
Although I am generally scientifically inclined, I would have to brush up on light detector physics and engineering to add to the quantitative debate. After reviewing various sensor performance curves with the tool linked to above, it looks to me like the sensor generations seem to be progressing to a flatter curve, but it is not clear to me whether that represents an improvement or not.
However, having put a few thousand frames of captures through my OMD, maybe the following highly subjective and aesthetically directed comments may provide useful input? Someone else may be available to explain to me why I am doing these things, and whether there are better choices.
My first inclination was to use the OMD at base iso whenever possible, because the colors come out absolutely gorgeous and if you ettr and botch it up there is headroom to recover quite a bit in pp (Aperture 3 recovery in my case). You could argue that specific good color examples may have also been due to good lighting, or good lens drawing, or post processing for optimal curves, but I have personally never been able to "nail" the color tones and ranges so easily with any camera previously. Admittedly I gave it a shot with Kodachrome and Velvia back in the day.
That said, when I get to iso 1600 and beyond, I start to think about converting to monochrome because the colors start to go wrong. I can still get reasonable results at 1600 and with NR up to ~4000, but it no longer looks natural, and if there is skin involved, a careful B&W treatment with virtual red or yellow filtering added keeps the image reasonably flattering if not ideal.
Now that I have digested this thread a bit, my recent and completely unrelated switching to auto iso with a limit of 400 or 800 in less than perfect lighting conditions makes sense a bit. Keep in mind that I am mainly an enthusiastic amateur working with ambient lighting, and only rarely supplementing it with fill flash or other sources. The reason I switched to auto iso was to still get good colors across the DR (shadow through highlights), but in less than ideal conditions. I have never felt the need for NR up to 400 or so with this camera after getting a reasonable exposure, and usually a light touch of denoising at 800 with moderate at 1600, and then nothing but B&W along with heavier NR above that. This upper end is probably the range of conditions that moved me to switch over to and then push process Tri-X 10 to 20 years ago.
Assuming the sensor at different gains keeps the Rs Gs and Bs looking the way my eyes expect them to be as the overall luminance drops, I am with the group that proposes to push iso whenever possible to get the histogram to push all the way to the right. If my rods and cones start to diverge with respect to what they expect to be seeing on the screen or print, and I start to see bias or artifacts rather than natural (or at least pleasing or useful) renditions of colors and textures, that triggers a different approach to the image in pp and a slightly different plan during capture – I tend to focus my attention on mid-tones when exposing for monochrome.
I have no idea whether the base iso vs higher iso comparisons proposed take any of these aspects of IQ into account, but you can be sure I will be following along closely to see if I can learn a way to wring out another eV or so of useful range out of this sensor.