I am WAY too lazy to look at the board. Here are some thoughts. The most orthodox way to deal with the trace widths is to choose a stackup and trace width so that you get the desired impedance (50 Ohms). For surface layer traces, the only thing that really matters is the distance to the ground (or power) plane and the trace width. The copper thickness has a small effect also. It is certainly possible to create a stackup such that a 10 mil trace has a 50 Ohm impedance. I have done this. You end up using a fairly small gap between signal and plane, and a fairly large gap between plane and plane, assuming that the board is 62 mils overall.
If Maxim didn't use 50-Ohm traces, there could be a few possible explanations. It could be because they don't know any better or don't care, or they could have something up their sleeve. I have to assume that they know what they are doing, so I'll rule out the notion that they don't know any better.
They might not care about the impedance if the traces are very short. Afterall, if a trace is only 0.05 lambdas, then it doesn't really matter if the impedance is right.
Another possibility is that they use traces which are a little bit too narrow (high impedance) to compensate for the parasitic capacitance of the pads. I don't know if this really works, but I have heard people advocate it before. If you are really interested, you can go find the formulas for pad capacitance, and model the traces as lossless transmission lines and simulate it using spice or something similar. You could do 16 simulations for 4 different capacitances, and 4 different trace impedances. It might be just as easy to calculate the input impedance using a spreadsheet, since you are only using one frequency.
T1 _______________
---+--------) )-----+-----+ = C1 --------------- = C1 | | | \ GND GND / R1 \ / | GND
This is the setup when power is coming in from a connector to a device on the board. C1 is the capacitance of the connector pad, C2 is the capacitance of the receiving device pad, and R1 is meant to model the input impedance of the receiving device. If you have reason to believe that the input impedance is not 50 Ohms resistive (for example if you have a spice model or listed package parameters), then replace R1 with a more accurate impedance, Z1. T1 is the trace modelled as a transmission line.
If you are transmitting off board, then you could just sort of turn it around.
Have fun.
--Mac