The wiggle is present on the inside track at corners (or the shorter overall) to equalise the track lengths of a differential pair - that is any two wires that use differential signalling to cary data. If the tracks were not the same length, the noise-cancelling benefit of a differential signalling would be lost.
While the physical-layer components of most modern LVDS signalling (PCIe, HDMI, DVI) include de-skew or 'elastic' buffers to compensate for differing track lengths between pairs, skew within a pair must be avoided with these physical layout techniques.
Following comments by OP:
Taking Gigabit ethernet as an example, as this might be more familiar to you: The CAT6 cable has eight wires, which if you tear open the outer insulating sheath are twisted together in pairs, so wires 1+2 are twisted together as pair one. Next to this lies pair 2, which is wires 3+4 twisted together, pair 3 comprises wire 5+6 twisted together etc. It's important to keep the pairs the same length, because they contain copies of the same signal sent with opposite polarities (one is positive, while the other is negative). If and only if the wires are the same length, the signals arrive together (given the fixed speed of electrons), which allows any common-mode electrical interference to be rejected in the magnetic coupling.
The four pairs themselves however do not have to be exactly the same length because the gigbit auto negotiation process calibrates the elastic buffers (and echo cancelling units) such that any minute discrepancies in arrival time are removed before the higher level components do their work.
The same thing is happening on this circuit board. The immediately adjacent/close circuit board traces are "the pairs" and are kept the same length to allow the differential receivers to reject noise, although electrically rather than magnetically. You can see the HDMI connector carries several such pairs, and no attempt is made to keep one pair the same length as the pair next to it ("between pairs"). There are however some limits in the size of the elastic buffers (in bytes) after which the cable becomes non-operative or downgrades. It would be fun to experiment and find the limits in millimetres.
This picture of a HDMI plug shows the differential pairs:
The 90º intersection will not be a issue for most traces. If your trace is very thin (< 15 mill) and/or the PCB boards aren't being professionally manufactured you can use mitered traces (chamfer) as helloworld922 pointed out. Adding chamfers will eliminate the 90º intersection and help to strengthen the trace. For high frequency traces (> 1GHz) it also helps reduce signal reflections.
The lines have constant impedance along their length regardless of
routing which is a highly desirable property. Note: this is less true
at the highest frequencies where abrupt corners in a track also cause
reflections. For this reason most PCB designs employ mitred (45°) or
chamfered (curved) corners.
There are other methods to help prevent signal reflections, like the following image shows:
The image above is from Microstrip, Stripline, and CPW Design.
For more reading material, take a look at 90 Degree Corners: The Final Turn.
Best Answer
1) Equalisation of length of pairs of traces
From Board Design Resource Center
2) Delay (e.g. of clock for timing purposes)?
See also Adding delay intentionally
3) Reduce signal reflections due to discontinuities in trace width?
from Circuit Board Layout Techniques
See also How should I lay out timing matched traces?