I don't quite understand myslef why it works differently. All the GPS related BLOBs are the same on both CM10.1 and 10.2: gps.omap4.so, glgps, rild and other ril libraries. I even compared the high level Java code between 10.1 and 10.2 to see if there is anything glaringly different but no, most of the code is similar. Well, 4.3 adds bunch of geo-fencing related stuff, but our GPS don't support it so the code is not used. The logs don't show much except that the glgps initialization and interactions between it and RIL (for AGPS) seem to be pretty much the same. But in the end on CM10.2 the log shows that glgps won't return location fix for a very long time and would loop returning empty data.
On the other hand I can't say that GPS doesn't work at all. Yes it's slower but I'm always able to get a fix. Not in 5-10 seconds as on CM10.1 but usually within a minute, even on a cold start. To me it's acceptable.
Btw as far as I remember I'd noticed some strange behavior on CM10.1 as well. Within just few seconds GPS Test would show a fix with 5-10 satellites, but the actual distance (accuracy) was not stabilized and would bounce around 500 feet for some time until it settles on usual 26-52 feet. Should not be happening is there really was a good fix.
The only time I was able to get a gps fix I waited 15 minutes then following fixes was quick until I rebooted my phone of course. This is a very annoying problem but thanks for your continuous work.