I think this is necessary to keep the documented model accuracy regarding mismatch, i.e. is very advisable for devices which have to match rather well - as in larger geometry processes, too. In these low size sub-micron processes this is getting more and more important, because of a relative increase of mask adjustment and lithography deviations, and - last not least - the erratic distribution of dopant atoms.
There seems to have recently appeared (I do not watch closely,
may not be -that- new) mask techniques revolving about what
they call "1-D layout". Idea being that if you constrain all poly
(say) to run N-S, and active E-W, you can do a better job at
lithography than allowing all kinds of orientation randomness.
The DSM nodes will take any help they can get.