I tried to build different oscillators with Citizen watch crystalls rated at 5 ppm. I built my own circuits with BJTs, I attached them to PIC, I did PCBs, breadboards. I tested them for weeks for time stability. Everything that I had built had great accuracy with errors less than a second a day. Therefore, I wouldn't expect such thing as PCB layout to have such a huge influence. May be the crystal is bad? Did you try the crystal from the previous PCB?
How about the temperature? Are they around 25C? They may be very sensitive. Here's the accuracy vs. temperature graph for Citizen crystalls.