Compare bitrate to track length and rise time.
While a minimum risetime is not specified, it's unlikely to be much less than, say, 1/3 the maximum; the RS-422 transmitter's 15ns max. is the most critical in this regard. 5ns at 67% speed of light is 1m, and we should be less than 1/4th this to avoid transmission line effects, or 25cm.
It is highly unlikely that 15cm of mismatched or uncoupled traces will run afoul of either of these interfaces.
The worst mismatch you can get is probably downward, using an obnoxiously wide trace giving an impedance of, say, 10s of ohms; this will basically manifest as lumped equivalent capacitance, which will still have little effect on the system; but will reduce bandwidth and signal quality if many such stubs are connected to the bus. (You can't make trace impedances very high, even if removing ground plane and using minimum trace width; typical widths on 2 and 4 layer PCBs are in the 50 to 150 ohm range.)
Not to say you shouldn't; and you certainly can. Simply to say, the advantage at these frequencies will be marginal. But it is good practice for working with faster interfaces where it is critical, and a marginal advantage is still an advantage.
Or, to put it more generally: a differential interface must always use differential traces with controlled impedance; the question is, how tight the impedance matching and coupling distance should be, over what lengths. For these standards, those lengths are approximately the length of a whole board, so it would be hard to actively abuse the signals to a significant degree. Likewise the allowable mismatch within that length could easily be, say, ±50%, which basically means, use the default track width and you're set.
The signals should still be routed as a pair, just so they are exposed to similar noise environments -- any noise picked up along the route can subtract out at the receiver. They don't have to be particularly close.
Just a general comment about the dI2C chips: notice they specify a 0...VCCB common mode range only. (Compare with RS-422/485* which specifies -8...+12V, suitable for most industrial applications.) Compared to the ~1.5V signal level, this will offer enough range to deal with commercial level noise sources (a few volts conducted/radiated), but not transient interference such as EFT and ESD. I don't see mention of transient immunity in the datasheet, other than to claim that it exists (it's mentioned, but without specifying any level and test setup, this is meaningless). I would recommend shielded cable for this link, particularly if you need to withstand harsher (industrial or worse) environments. (The shield may in turn require some means of dealing with ground loop; the shield can be RF-bypassed with capacitors at one or both ends, for example.) Perhaps they have better immunity than this, I don't know, but it is unfortunate that they didn't design for a wider input range, or specify immunity directly.
*Curiously, I don't see an input common mode range specified for the MAX307x series. It's probably fine, but I would be tempted to shop around for an equivalent part that does specify this -- just to be sure.
