cdwilson, please take on all your questions:
... Bob Smith termination ... may not be ideal (or correct). If this is the case, why is Bob Smith termination still recommended in Ethernet design guides from Intel, TI, etc?
There is no conspiracy here because it's not the case. Auto-crossover (Auto MDI-X) function is popular now (and was popular even in 2000s) and it requires a symmetric CMC enabled transformer. Again, as Royce Bohnert shows, CMC solves the problem in its root making the need of the termination practically negligible. Also, CMC enabled transformer make possible to use various mediums with various impedances in the single design: for example, 100BASE-TX is intended to operate over 100-Ohm UTP and 150-Ohm STP (by IEEE Std 802.3). Repeating, there is no conspiracy and/or irrationality here.
Is the author of the article incorrect in his analysis?
IMO, the author looks at the problem narrower than it is: there are many variants of BS termination (e.g. Intel'2001, for another ones lurk by yourself :-) in the first, and the termination is intended to solve not only EMI but also ESD issues in the second. Looks over the past 15-20 years (with help of Google, of course :-) and you'll see how the termination evolves.
Or do people just not care that the return loss could be improved with a proper choice of resistor?
They care. But also they care about the cost (and other features of the design) in complex.
... do the magnetics always come with chokes built in?
CMC is a common feature now. During my 10+ yrs carrier i used only CMC enabled magnetics. Maybe it is because i like AutoMDIX :-) You could browse Pulse, PCA, Halo, and many others to make your own opinion for today.
If the proper termination was used, would it be possible to use magnetics without the chokes?
If you find such a transformer you could, but i think it is problematically now :-) PHY chip manufacturers (Micrel/Kendin, Intel/LevelOne, TI/National) recommend CMC enabled ones and specifies corresponding BS termination variant based on that.
Also, the Bohnert presentation addresses emissions but I'm also curious whether this applies the same to immunity? (i.e. imbalance in the PHY receiver can reduce tolerance to immunity)
Yes, it applies: less emissions -> less crosstalks/disturbances/imbalance. Bohnert states:
Impedance is not constant along length of cable. Different twist rates on each twisted pair cause cancellation of field every few centimeters.
If it's not about immunity, what is it about?