I am well aware of the matters you mention, but as I first wrote, and to quote a paragraph from the article you linked:
"Despite their external sensor differences, both headsets utilize magnetometers, gyroscopes, and accelerometers built directly into the headsets to handle tilt tracking."
Now, these IMU units are not for tilt tracking - the writer got that slightly backwards; The optical tracking is perfectly capable of giving you the orientation of devices - not just their position.
Instead, they are for filling in, smoothing out, adding resolution to-, and improving dead reckoning for- (inherent from improving temporal resolution), the "blank spaces" between optical tracking updates. I am sure you are familiar with the process of the term "sensor fusion", for putting the data from the different systems together.
...And the amount of data that is transmitted to the host from them (the IMUs) greatly overshadows that which comes from the photoreceptors.
The quote also says: "headset", specifically, but the same goes for the controllers, and any other tracked object that needs a decent degree of fidelity. With only LEDs or photoreceptors (...and both are arranged in similar "constellations", of course - it's just a matter of which end do the watching), you will get more laggy and jerky tracking, that is more susceptible to intermittent occlusion -- good enough for a more or less static object, like say a keyboard, or teacup, or furniture, but for controllers, you definitely want better,
The difference in play area size is due to a number of reasons, most important being: 1) that the Constellation tracking cameras have a limited field of view (...and focal range), which could be expanded with a different lens, but that would come at the cost of ineherently widening the field of view of each discrete camera pixel, too, reducing: 2) resolution. The latter you do get more out of than the absolute pixel count, on account of comparing neighbouring pixels, in order to position the light blob with subpixel precision, but first one needs to deal with sensor grain and other noise.
...and, again as said before, and however minor an issue this is; The driving of the constellation LEDs in not entirely passive. They flash, in groups, and synchronised to the cameras, between a high and low intensity light output level, so that these groups can be identified on the pictures that the host computer receives from the cameras. This may take very simple logic and small transistors, but logic none the less. The LEDs also need to be bright enough to not be drowned out by sunlight.
Lighhouse has its own set of technical weaknesses, but that is another story. :7
(EDIT: ...and as far as I know, none of the sensor fusion, nor trigonometry, is done on the Vive wands, at this time: Just like with the Touch controllers, IMU data is transmitted raw, for the host to deal with, and the same goes for the added data from the photoreceptors; Just their raw timing data (which translates not quite directly to angles). If it had been calculated on-device, that would mean significantly less data traffic, albeit at a cost of additional (relatively light) on-device processing.)