Firstly, 260 km off the coast is international waters, so I think all our discussion below's moot. I think ISRO's just being nice. Mea culpa.. I should have clicked through and read the link posted by input~2.
Having said that though..
It is possible that this could have something to do with reliability - or in the GSLV's case, lack thereof. To date, more GSLVs have ended up in the Indian Ocean than the correct orbit, its success rate is less than 30%, and it hasn't had a successful launch in the last four attempts. The PSLV by contrast has had 21 consecutive successful launches with around a 90% success rate overall. So maybe they just realise this one is more likely to fail.
I thought the launch azimuth was decided and corresponding NOTAMs detailing ZOEs were issued
regardless of predicted launch vehicle success or failure... and even assuming an FTS malfunction[2], preventing destruction..
i.e. if you're outside the boxes, you're always safe[1] - whether the rocket breaks up or not... (the cross range dimensions of the box are calculated based on the maximum deviation that the rocket can make - pitch and yaw without aero-structural loads ripping it to shreds, as well as the total burn time of the propellant).
And in any case, that doesn't answer the PSLV question. It's proven its reliability, so we shouldn't hesitate from sending it over Sri Lanka. That's more mass to polar orbit, almost for free! No design changes.. nothing..
Of course, this might still not be desirable if the spent stage would fall on to the Sri Lankan land mass (or even their offshore -
useful EEA) in the course of a nominal launch. If it would require us to clean up the surrounding area, it would probably nullify the cost advantage (but we still keep the greater mass to orbit advantage).
Now it's a question of numbers. How much does the launch angle and the subsequent yaw manoeuvre cost us in terms of mass penalty? Would it open up a new market segment?
[1] To the same extent that people around the launch pad itself are safe...
[2] On its own, and assuming the guidance input is all fine, this is probably ok to analyse. But if you couple a software error - like mixing up East and West... it gets ridiculous real fast.. lol