Well I guess it is what medved described. I have one 18w osram that does it daily (the lamp runs 12h per day everyday) and it has done it almost three years now. One t12 lamp in my hallway does it some days but since it's thicker it's hard to notice. Some other daily used lamps does it too, one of them being pl-s 11w.
why does it appears to happen always at the same end of the lamp?
At first I suspected it was inflicted by some interence coming from a power grid because I had not noticed it happening this frequent at my previous apartment.
On the mains frequency ballast the arc cathode root works in that way only 8.3 or 10ms (60 or 50Hz), then the other half cycle it cools down and only the next half cycle it has the next chance to become the root again. But during that time it cools down and so another spot may become the preferred root. When there are irregularities in the mains (e.g. some high current nearly saturating load and long distance to the substation; it may be asymmetric, mainly when permanent magnets are involved), the arc current may vary in shape and so alter the root spot condition just before the zero cross, so altering it's chances to become root the next cycle.
Personally I've seen such dancing to appear only sometimes (one day it was flickering, the other not; many times it starts to flicker without any apparent reason and after some time stops again, really random) and it didn't appear to depend on the lamp age or wear...